Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Feeding the machine: Challenges to reproducible predictive modeling in resting-state connectomics

Feeding the machine: Challenges to reproducible predictive modeling in resting-state connectomics PERSPECTIVE Feeding the machine: Challenges to reproducible predictive modeling in resting-state connectomics 1,2 3,4 1 3,5 Andrew Cwiek , Sarah M. Rajtmajer , Bradley Wyble , Vasant Honavar , 1,2 1,2 Emily Grossner , and Frank G. Hillary Department of Psychology, Pennsylvania State University, University Park, PA, USA Social Life and Engineering Sciences Imaging Center, Pennsylvania State University, University Park, PA, USA College of Information Sciences and Technology, Pennsylvania State University, University Park, PA, USA Rock Ethics Institute, Pennsylvania State University, University Park, PA, USA Institute for Computational and Data Sciences, Pennsylvania State University, University Park, PA, USA Keywords: Machine learning, Classifiers, Predictive modeling, Brain networks, Clinical an open access journal neuroscience ABSTRACT In this critical review, we examine the application of predictive models, for example, classifiers, trained using machine learning (ML) to assist in interpretation of functional neuroimaging data. Our primary goal is to summarize how ML is being applied and critically assess common practices. Our review covers 250 studies published using ML and resting-state functional MRI (fMRI) to infer various dimensions of the human functional connectome. Results for holdout (“lockbox”) performance was, on average, ~13% less accurate than performance measured through cross-validation alone, highlighting the importance of lockbox Citation: Cwiek, A., Rajtmajer, S. M., data, which was included in only 16% of the studies. There was also a concerning lack of Wyble, B., Honavar, V., Grossner, E., & Hillary, F. G. (2022). Feeding the transparency across the key steps in training and evaluating predictive models. The summary machine: Challenges to reproducible predictive modeling in resting-state of this literature underscores the importance of the use of a lockbox and highlights several connectomics. Network Neuroscience, methodological pitfalls that can be addressed by the imaging community. We argue that, 6(1), 29–48. https://doi.org/10.1162/netn _a_00212 ideally, studies are motivated both by the reproducibility and generalizability of findings as well as the potential clinical significance of the insights. We offer recommendations for DOI: https://doi.org/10.1162/netn_a_00212 principled integration of machine learning into the clinical neurosciences with the goal of Supporting Information: advancing imaging biomarkers of brain disorders, understanding causative determinants for https://doi.org/10.1162/netn_a_00212 health risks, and parsing heterogeneous patient outcomes. Received: 12 February 2021 Accepted: 8 October 2021 Competing Interests: The authors have declared that no competing interests exist. BACKGROUND Corresponding Author: In settings where large amounts of well-characterized training data are available, modern Frank G. Hillary fhillary@psu.edu machine learning (ML) methods offer some of the most powerful approaches to discovering regularities and extracting useful knowledge from data (Bishop, 2006; Goodfellow et al., Handling Editor: Olaf Sporns 2016; Hastie et al., 2009; Mitchell, 1997). Of particular interest are algorithms that, given a data set of labeled samples, learn a predictive model, for example, a classifier, for labeling novel samples drawn from the same distribution as the training data. Programs for training Copyright: © 2021 such classifiers typically optimize a desired objective function on a given set of training sam- Massachusetts Institute of Technology Published under a Creative Commons ples. Advances in ML have revolutionized the design of systems for natural language process- Attribution 4.0 International ing (Manning et al., 2014; Mikolov et al., 2013; Turian et al., 2010), computer vision (Bradski (CC BY 4.0) license & Kaehler, 2008; Deng et al., 2009; Forsyth & Ponce, 2002), network analysis (Hamilton et al., The MIT Press Predictive modeling and network neuroscience Classifier: 2017), and bioinformatics (Baldi et al., 2001; Larrañaga et al., 2006; Min et al., 2017). A num- An algorithm designed to classify two ber of publicly available ML libraries (e.g., Scikit-learn, TensorFlow) can now be deployed, or more groups using a given set of permitting “off-the-shelf” application of these analyses for a number of data types including variables. behavioral, genetic, and imaging data (Abadi et al., 2016; Abraham et al., 2014). Training: In one sense, predictive models trained using ML are like traditional statistical models, for A generally iterative process wherein example, regression: there are covariates, an outcome, and a statistical function linking the an algorithm is refined to better covariates to the outcome. But where ML algorithms add value is in handling enormous num- classify a subject into their correct bers of features or predictors, heterogeneous data types (e.g., images, text, genomic sequences, group by tuning the parameters for molecular structures, networks, and longitudinal behavioral observations), and combining selection of important variables. them in complex, nonlinear ways to make accurate individualized prediction, that is, a clinical Features: diagnosis. This review examines the use of predictive models in ML and resting-state connec- The variables input into the algorithm tomics with focus on several particularly important issues, including “overfitting” and its for use in classification. related consequences, sample size and implications for modeling clinical heterogeneity, and methodological transparency. Overfitting: When the algorithm is too closely Prediction Modeling in the Neurosciences attuned to the data it was trained on, to the detriment of the algorithm’s There has been growing use of ML to determine if brain network metrics can serve as classi- generalizability to new samples. fiers of brain disorders with several high-profile reviews recently published (Bassettetal., 2020; Braun et al., 2018; Parkes et al., 2020; Vu et al., 2018). Many of the canonical networks identified in rsfMRI studies (e.g., default mode network) have been of critical focus in studies of large-scale network plasticity in a range of brain disorders including schizophrenia (de Filippis et al., 2019; Lefort-Besnard et al., 2018; Progar & May, 1988; Steardo et al., 2020), autism (L. Chen et al., 2020; Glerean et al., 2016; Hegarty et al., 2017), Alzheimer’s disease and related dementias (Langella et al., 2021; Pellegrini et al., 2018; Salvatore et al., 2015), and brain injury (Bonnelle et al., 2012; Caeyenberghs et al., 2017; Gilbert et al., 2018; Roy et al., 2017). Dimensionality: While the high dimensionality of functional imaging data—relationships between hundreds The number of features given to the or thousands of time series observations—may push the limits of traditional modeling, ML algorithm. approaches can capitalize on the complexity of multimodal datasets (Baltrušaitis et al., 2019; Gao et al., 2020; Guo et al., 2019) and provide opportunity to examine interactions among variables otherwise impossible to test. Therefore, there is evident potential for the application of ML to incorporate a wide array of data structures into prediction modeling including behavioral, brain imaging, physiological measurements, and genetic markers. Growing Pains in ML and Resting-State Connectomics Perhaps the most common methodological concern in applied ML is overfitting, or training an algorithm to predict with very high accuracy features within a single dataset at the expense of predicting a phenomenon more generally (Dietterich, 1995; Ng, 1997; Roelofs et al., 2019; Srivastava et al., 2014). Overfitting has profound implications for reproducibility, portability, and generalizability of findings. Importantly, the difficulty of preventing overfitting is underap- preciated, and even typical remedies, such as cross-validation, can allow for analysis hyper- Cross-validation: A process to limit overfitting through parameters to become tuned, or “overhyped,” to a specific set of data (Hosseini et al., 2020; repeated splitting of the data into Poldrack et al., 2020). These concerns underscore the need for greater transparency in model training and testing sets to prevent selection, enforcement of model parsimony, and rigorous testing and validation of trained models overfitting. on independent validation data, with attention to class imbalance in the data, relative costs of false positives versus false negatives, and the tradeoffs between them (Varoquaux et al., 2017). Related to overfitting are concerns about the size or heterogeneity in the training and test samples (Poldrack et al., 2020). When a sample is overly restrictive along dimensions that Network Neuroscience 30 Predictive modeling and network neuroscience influence outcome in neurological disorders (e.g., severity of disease, age of onset), it may reduce the study reproducibility and the ability to predict the phenomenon as it naturally occurs (Caruana et al., 2000; Hawkins, 2004; Schaffer, 1993; Srivastava et al., 2014). As an example, an investigator may have access to a large database of cases of individuals diagnosed with a neurological or psychiatric disorder that can be used for training and test datasets. Even with conservative training and only single exposure to the testing dataset (the gold standard), the result may not generalize if the sample is restricted in its range of characteristics with respect to demography, symptom severity, or disease/injury chronicity. Goals of this Review There is significant and justified enthusiasm for using ML approaches to advance our under- standing of brain disorders. With the ever-increasing application of ML in the study of resting- state connectomics, the importance of the implementation of and adherence to best practices is further underscored. Given this backdrop, we review 250 papers using ML for diagnosis or symptom profiling of brain disorders using resting-state fMRI methods, coding information regarding the methods used with particular focus on how algorithmic “success” was deter- mined, the use of a lockbox dataset (i.e., a data set that can be accessed only once at the Lockbox: A set of data intentionally set aside end of the analysis, also called a holdout set, a test set, or an external set), transparency in before training the algorithm and the approach, sample size and heterogeneity, and the types of conclusions drawn. We aim used exactly once after training to to provide a summary of the state-of-the-art in ML applications to one area of clinical neuro- test the generalizability of the result. science with the goal of identifying best practices and opportunities for methodological improvement. While we focus on resting-state fMRI connectomics here, the issues addressed likely have relevance for a wider range of ML applications in the neurosciences. Method: Literature Review We conducted a literature search using the following search terms in the PubMed database: (ML OR classifier OR supervised learn OR unsupervised learn OR SVM) AND (brain) AND (network OR graph OR connectivity) AND resting AND (imaging) AND (neurological OR clinical OR brain injury OR multiple sclerosis OR epilepsy OR stroke OR CVA OR aneurysm OR Parkinson’s OR MCI or Alzheimer’s OR dementia OR HIV OR SCI OR spinal cord OR autism OR ADHD OR intellectual disability OR Down syndrome OR Tourette) AND “humans”[MeSH Terms]. We did not bound the date range for our search, but we excluded non-English papers, review papers, and animal studies. We also excluded papers that were based on simulations or other nonhuman data. Our initial search returned 471 papers that were reviewed for inclu- sion. Two reviewers independently screened all of the papers returned from the above search at the title and abstract level for exclusionary criteria. By examining each paper title and abstract, papers were excluded based on the following categories: (1) examined structural brain imaging only (n = 98; 21%); (2) did not examine a clinical phenomenon (n = 59; 13%); (3) focused on automated tissue segmentation or lesion identification (n = 48, 10%); (4) was focused on algorithm or method development without clinical diagnostics (n = 41, 9%); (5) used other imaging approaches such as EEG/MEG (n = 33, 7%); (6) did not implement formal network analysis (n = 27, 6%); (7) was not an empirical study, including reviews and perspectives (n = 25, 5%); (8) did not use machine learning (broadly defined) or classification (n = 13, 3%); or (9) another reason consistent with the exclu- sionary criteria (n = 9, 2%). This resulted in exclusion of 353 papers, and for the remaining 118 papers (25%) the full paper was included in the final analysis. For the full-text review, two Network Neuroscience 31 Predictive modeling and network neuroscience reviewers were assigned to each specific section based on their respective specialties and completed a full analysis on multiple papers to identify any potential inconsistencies between the reviewers. Following this brief training for inter-rater consistency, the reviewers completed a full analysis of the papers independently. Based on feedback during the review process, we broadened our review to include terms sensitive to papers using deep learning approaches. A second identical keyword search to the above was conducted, while inserting the following terms to capture ML and deep learning approaches: “(deep learn* OR deep belief network OR multilayer perceptron OR autoencoder OR convolution neural network OR artificial neural network OR generative adversarial net- work OR machine learning OR ML OR classifier OR supervised learn OR unsupervised learn OR SVM) AND …”). The second search (April 2021) revealed 625 papers and based on abstract review (or full manuscript review if necessary), 405 papers were excluded based on the following categories and several for multiple reasons: (1) did not use machine learning (broadly defined) or classi- fication (179, 28.6%); (2) did not examine a clinical phenomenon (n = 90, 14.5%); (3) did not implement formal network analysis (n = 29, 4.6%); (4) used other imaging approaches such as EEG/MEG/PET (n = 28, 4.4%); (5) reviewed already existing literature, no new analysis (n = 24, 3.8%); (6) fMRI data were not included for prediction modeling (n = 22, 3.5%); (6) analysis included structural neuroimaging only (n = 12, 1.9%); (7) prospective study or proposal (n =6, .009%); (8) study not available in English (n = 3, .004%); (9) animal studies (n = 2, .003%); and (10) other reasons consistent with the exclusionary criteria (e.g., pilot studies, lesion segmen- tation studies, n = 11, .018%). This resulted in retention of 220 papers from our second search (n = 625). After eliminating redundancies with the outcome of the initial search (n = 471, n = 118 included), the final review included 250 unique papers for analysis. A flowchart for the literature review is provided in Figure 1. Data Coding To understand the methodological factors shaping machine learning use, the type of classifica- Permutation testing: tion algorithm utilized, subject population count, and the use of permutation testing with blind A method for testing the final feature analysis, as defined by Hosseini et al. (2020), were collected. Additionally, key information per- set against chance performance taining to the description of features input into the algorithm, the classifier design, and the per- through repeated randomization of formance reporting metrics chosen to measure chosen ML technique’s findings were collected. class labels (i.e., patient vs. healthy In addition to the year of publication, specific demographic factors of the participants used in control) and comparison of the distributed accuracy to the observed each paper were recorded. These factors include age, years of education, handedness, age of performance. diagnosis (where applicable), and socioeconomic status. Features used to train the algorithm were recorded including the use of network metrics, behavioral data, injury or disease charac- teristics, genetic information, blood biomarker information, medical history, and demographic factors. For network metrics specifically, information regarding the node definition and count, edge definition, and whole-brain versus subnetwork analysis were additionally recorded. Elements shaping the result reporting of the classifier, including the metrics chosen by the article, the type of cross-validation technique, ablation reporting, and use of a lockbox (i.e., a strictly observed separation between data used to train/optimize the analysis parameters and data used to assess generalizability; see Hosseini et al., 2020) were a primary focus of this review. Because classifier accuracy was a focus for our review, this was coded for all papers, and in the case of multiple analyses, the test with the highest performance at the most stringent Network Neuroscience 32 Predictive modeling and network neuroscience Figure 1. PRISMA flowchart of literature review. *An initial PubMed search was conducted, following valuable feedback, an updated search was conducted including articles up to the year 2021, and which included terms to broaden the search to include deep learning algorithms. For details, please see section Method: Literature Review. **Initial Review did not delineate removal at particular step; updated review includes a step-by-step workflow. ***220 from updated search + 30 nonduplicates from initial search. Modification of flowchart provided by Page et al. (2021). validation stage (cross-validation or lockbox) was selected for analysis. In cases where papers did not report total accuracy, but did report sensitivity and specificity, we calculated an accu- racy value based on sensitivity, specificity, and sample size (see Supporting Information For- mula: Formula S1). The presence of a lockbox was determined through keyword search of the full text for terms such as “holdout,”“external,”“test set,”“testing,”“withheld,” or “validation-set,” followed by a critical reading of the methodology. To qualify as a lockbox, the article had to (1) set aside a subset of data for the purpose of testing the algorithm performance following training, (2) make explicit that no part of the data in the lockbox was included at any point during algorithm development, and (3) not report multiple training/testing phases to arrive at the final lockbox performance. From the 250 papers, 44 (16.8%) included a test dataset, and of those, 32 included both lockbox and cross-validation performance reports. Network Neuroscience 33 Predictive modeling and network neuroscience Interpreting Model Performance Investigators have several tools at their disposal to better understand the impact of individual features on the final performance of the algorithm. While there are myriad ways in which spe- cific tools can be implemented, we searched for and coded the four most common methods observed in this literature set listed here in order of commonality: (1) feature importance, (2) Ablation analysis: permutation testing, (3) ablation analysis, and (4) consensus analysis. A measure of the contribution of Feature importance, or the discriminative power of an individual feature as assigned by the particular variables through manual trained algorithm, is an inherent element of many machine learning methodologies wherein removal or addition of specific features during training. features are ranked by their relative impact on the decision boundaries set by the algorithm. An article was coded as including feature importance if it included a report of some or all top- ranking features with some quantitative analysis of their relative contribution, such as (but not limited to) Gini index, Kendall’s tau values, or the correlation coefficient r. Permutation tests use thousands of randomized shufflings to simulate the distribution of pos- sible outcomes that a given comparison could have revealed if the independent variable was meaningless with respect to the analysis (i.e., the null hypothesis distribution). This technique can then measure the likelihood of an observed analysis outcome with an observed set of data or analysis outcome. Papers that run such analyses and report the likelihood of chance performance, generally in the form of p values, were coded as reporting this valuable analytical technique. An ablation analysis examines the performance of the algorithm when portions of the algo- rithm are removed in order to either improve performance (i.e., during training) or to determine which portions of the algorithm or dataset contribute to the algorithm’s accuracy. This is sim- ilar to feature selection in the context of neuroscience (Guyon & Elisseeff, 2003). For a paper to demonstrate an ablation report per our coding scheme, it must show the changes to perfor- mance in training, whether as a function of feature reduction or of iteration count. Consensus analysis is another common technique for analyzing relative importance of fea- tures by way of the ratio of times a feature is selected across the number of training/validation folds. Articles providing either a raw count or some other form of occurrence frequency for key features were coded as demonstrating a basic consensus analysis. RESULTS Representation of Clinical Disorders in Review The final review included 250 studies largely composed of case-control designs focused on prediction modeling of diagnostic accuracy. The studies ranged from traditional neurological diagnoses (Alzheimer’s disease, brain injury) to psychiatric disease (depression, anxiety), to neurodevelopmental diseases (schizophrenia, autism spectrum). A summary of the distinct clinical disorders represented in the review is provided in Supporting Information Table S1. The largest representation of studies examined Alzheimer’s disease or related dementias (n = 66, 26.4%), depression/anxiety (n = 40, 16.0%), schizophrenia spectrum disorder (n = 34, 13.6%), Autism spectrum disorder (n = 33, 13.2%), and brain injury (n = 15, 6.0%). For training, sample sizes ranged from 17 to 1,305 subjects for total samples and 8 to 653 for group-level data (case-control designs). For test datasets, the sample sizes for the total pop- ulation ranged from 8 to 477 subjects and group-level data ranged from 1 to 185. See Table 1 for breakdown of training and test dataset sample sizes based on population and subgroup. These sample sizes are consistent with those observed elsewhere (Poldrack et al., 2020), and we anticipate that the large majority of studies present in this review were underpowered Network Neuroscience 34 Predictive modeling and network neuroscience Table 1. Sample sizes for population and subgroups in training and test datasets Training set (n = 250) Test set (n = 44) Sample Total Subgroup Total Subgroup Range 17–1305 8–653 8–477 1–185 Mean 126.7 50.0 96.6 38.1 Median 77 29 39 20 Studies with n ≤ 50 80 (32.0%) 192 (76.8%) 23 (52.3%) 35 (79.6%) Studies with n ≤ 30 24 (9.6%) 136 (54.4%) 14 (31.8%) 28 (63.6) Studies with n ≤ 20 3 (1.2%) 82 (32.8%) 8 (18.2%) 22 (50.0%) for reliable prediction modeling, resulting in low confidence in the portability of the reported algorithm and reproducibility of the finding in other samples. Network Characteristics Consistent with the inclusionary criteria, 100% of the studies used at least one network metric as input during classifier identification. Table 2 provides descriptive data for the types of network studies included and the characteristics of the networks analyzed. A majority of the studies used whole-brain network information as features (73%). Similar to other exami- nations of the use of network neuroscience to examine clinical disorders, there was a wide range of brain parcellation values, resulting in graphs of widely varying sizes and complexities (Hallquist & Hillary, 2018). Sample Characteristics Sample characteristics including demographics and common clinical indicators were exam- ined. While age of the sample was commonly reported, only 25.6% of studies included a mea- sure of time since diagnosis, fewer still reported age of diagnosis (10.8%), and few included demographic factors such as race (5.6%). Several studies lacked a “healthy control” group. In these cases, the studies either compared the same sample at two timepoints (1) or classified Table 2. Network data: Characteristics of functional brain imaging network analysis including in prediction modeling Range Median Mean (SD) Mode Network Nodes (parcellation) <10 to 67,955 90 483.9 (6,654.5) 90 n = 221* Correlation Partial Causal (e.g., Pearson’s r) correlation Multiple modeling Other Edge Definition n = 247* 67.9% 3.2% 6.1% 3.6% 18.3% Whole brain Modules/subnetworks Nodes/seed-based Unclear Scope of study n = 250 73.1% 19.0% 7.9% 3% * Note: All studies included defined nodes, but in some cases the exact number of nodes was unclear with respect to ML training (n = 30). Similarly, all studies examined connectivity between brain regions, but for a small number of studies there was no clear edge definition (n = 3). Network Neuroscience 35 Predictive modeling and network neuroscience against two separate clinical groups (5). See Supporting Information Table S2 for details regarding reported demography. Classifier Information Critical to our goal was to assess the classifiers used in this literature, the most common input features, and how the classifiers were evaluated. Table 3 provides summary information for characteristics of classifiers used across the 250 studies. Support vector machines were the most prevalent ML algorithm selected, appearing as at least one of the utilized classifiers in 68.4% of papers. The three next most common techniques used were linear discriminant analysis (8.8%), regression classification (8.8%), random forest (8.0%), and deep learning approaches (8%). In the papers reviewed, 18.8% implemented multiple ML algorithms. The feature types chosen to be used for a classifier, while not inherent to the quality of the ML algorithm, do speak to the nuances of the aims of the collected studies. While every study collected some form of injury or disease characteristic (at the very least for the purpose of identifying patient and healthy control groups), roughly 8% of studies included some form of these metrics as features to include in the classifier, and even fewer included demographic information (7%) and/or behavioral or cognitive data (4%) as features for training. Medication history, current medications, or other clinical metrics were rarely included as features (<1%). Only one study utilized blood biomarkers, and none included genetic factors in addition to network metrics, revealing an as-of-yet untouched potential for more nuanced feature-pattern discovery. Accuracy was the primary performance metric, with roughly 86.8% of papers reporting accuracy as the baseline measure of performance. More than two-thirds of studies included sensitivity and specificity, two metrics vital for identifying bias in classification, especially in the case of sample imbalance, whereas only 40.0% of studies included a full area under the receiver operating characteristics curve (AUROC) report; 12.0% of studies included predictive power, and 12.0% of studies included some form of regression analysis in addition to the Table 3. Classifier types, inputs, and metrics for evaluation during classification Deep Classifier SVM RF KNN LOG_R LDA learning Multiple Other Frequency* 171 (68.4%) 20 (8.0%) 17 (6.8%) 22 (8.8%) 22 (8.8%) 20 (8.0%) 46 52 (20.8%) (18.0%) Brain Injury/ Behavior/ Genes/ Inputs into network disease cognitive Medical blood classifier metrics actor Demographic data Hx Meds biomarkers Other Frequency 100% 13.5% 10.1% 5.9% 2.5% 1.7% 0% 1.6% Metric for AUC Predictive Regression evaluation Accuracy Sensitivity Specificity (AUROC) power outputs Other (e.g., F1) Frequency 87% 70.4% 69% 40% 12% 12% 20% Note: SVM, support vector machine; RF, random forest; KNN, k nearest-neighbor; LOG_R, logistic regression; LDA, linear discriminant analysis. *Total >100%, including studies with more than one classification approach. Network Neuroscience 36 Predictive modeling and network neuroscience Table 4. Validation measures Validation procedures Yes No Unclear Cross-validation 94.1% 4.2% 1.7% Lockbox 20.3% 79.7% 0.0% If lockbox, compared 70.8% 12.5% 16.7% once (n = 24) classification outputs. Finally, 20.8% of studies utilized some other form of metric performance reporting, such as F1 scores; all such measures fitting the “other” category were utilized in less than 5% of papers. Validation Approaches Most studies utilized some form of cross-validation, including leave-one-out cross-validation (LOOCV) (58.8%), k-fold (35.6%), nested approaches (11.2%), and multiple approaches (9.2%). Of note, 12 (4.8%) of the studies did not report any cross-validation use. In these cases, the authors either provided no alternative validation method (n = 8) or used a lockbox but no cross-validation (n = 4). The key diagnostic for overfitting, use of a lockbox, was only utilized in 16.8% of studies (Table 4). Of the studies using a lockbox, 81% (34/44) made clear that iterative training never permitted access to the test (lockbox) data, and 73.8% (31/44) reported accuracy results for both the training and lockbox data. Interpreting Model Performance Feature importance measures were the most common metric included, with nearly half of all studies including some level of quantitative analysis (47.2%). The other three common tech- niques for model interpretation were observed at a rate ranging between 1-in-3 to 1-in-5 papers. Permutation testing was included in 34.0% of all studies. Ablation reports were included in 27.7%, and consensus analyses were utilized in 20.0% of all studies (see Table 5). It was rare for examiners to include some form of all four approaches described here (2.8%), but about one-third of papers integrated two to three techniques (35.2%), more than a third integrated at least one method (38.4%), and finally one-fifth of papers did not conduct an analysis of feature importance (22.8%). Table 5. Common techniques for enhancing model interpretation Model interpretation techniques Yes No Feature importance 47.2% 52.8% Permutation testing 34.0% 66.0% Ablation analysis 27.7% 72.3% Consensus features 20.0% 80.0% Note: >100% due to multiple approaches used in some studies. Network Neuroscience 37 Predictive modeling and network neuroscience Figure 2. A histogram of accuracy scores for n = 250 studies reviewed reveals distinct distributions and median scores (organized in text boxes by color) for classification accuracy based on results using no validation, cross-validation, and external validation (i.e., lockbox). Classifier Performance Measuring ML performance with no form of cross-validation or lockbox validation produced a median accuracy of 97.1%. ML application using a cross-validation produced a median clas- sification accuracy of 86.8%. When classification was performed on lockbox data, the median classification accuracy dropped to 73.9%. The distribution for accuracy values across these distinct cross-validation approaches is reported in Figure 2. DISCUSSION While our review confirms the exciting promise of ML approaches in the network neurosciences to advance overall understanding of brain disorders, there also appears to be room for method- ological growth. We first make several observations regarding clinical sampling and how net- work neuroscience has been implemented in this literature as inputs for predictive modeling. We then focus the remainder of the discussion on critical issues that, if addressed, can bring greater precision to the use of ML in the neurosciences and ideally accelerate our understanding of the pathophysiology of brain disorders. In the following we highlight several issues in order to foster discussion in the literature: (1) need for uniformity in the creation of neural networks for predic- tion, (2) issues of sample size and heterogeneity, (3) need for greater transparency of methods and reporting standards, (4) the focus on classification accuracy at the expense of other informa- tion, and (5) explainability and feature importance. We outline these concerns and link them to eight decision points in the typical ML processing stream outlined in Figure 3, which serves as a roadmap for key considerations and reporting opportunities at each step of the training process with the goal of improving the interpretability, reproducibility, and clinical utility. Network Neuroscience 38 Predictive modeling and network neuroscience Figure 3. Illustration of distinct decision points in the typical ML pipeline in the papers included in this review. We identify eight distinct decision points where there are opportunities to report (R) information to maximize transparency. R1a: Justify classifier model choice from previous literature, limitations of data, and clinical goals of study. R1b: Explain how data were split between training and test sets (i.e., lockbox), including sample sizes and any matching of demographics or disease variables. R2: Make clear decisions about how the network was created, including edge definition and brain parcellation. R3: Make explicit the specifics of the model (e.g., parameter settings, kernel functions). Make clear which features (e.g., network metrics, clinical variables) are included in the model. R4: Report cross-validation method selection and implementation; justify use in context of sample size and potential risk of performance overestimation. R5: Explain the conditions necessary to terminate algorithm training, such as target performance or minimal feature count. R6:Make explicit the hyperparameter settings and any manual tuning of parameters between training itera- Feature weight: tions. R7a: Report training set results, including model performance, feature weights, and feature The discriminative ability of a given counts across training iterations. R7b: Explicitly state that preprocessing is unchanged from the final feature as measured and quantified algorithm derived from training and that during training there was no access to the lockbox; provide through various methodologies. the final averaged cross-validation performance and feature importance for the test set. R8: Provide clear interpretation and explainability for the model by highlighting any key findings in context of potential clinical utility (i.e., relevant regions of interest’s connectivity patterns). Network Neuroscience 39 Predictive modeling and network neuroscience Sample Sizes and Clinical Heterogeneity Roughly one-third of the studies sampled in this review had no more than 50 subjects in their total sample size for use of training and internal validation of their results. Furthermore, half of all lockbox sets examined had subgroup sample sizes of 20 or less. Thus, roughly half of the studies reviewed were likely underpowered to capture the stage, severity, and symptom con- stellation evident in heterogeneous neurological and neuropsychiatric disorders. Moreover, small samples likely contributed to the use of LOOCV (58.8%) instead of k-fold (35.6%), which may be more representative of the dataset (Poldrack et al., 2020). Clinical characteristics of the participants (representativeness) that comprise a sample may be just as vital as the sample size. Most neurological disorders maintain heterogeneous pre- sentations. For example, over a quarter of the studies focused on either schizophrenia or autism, both understood as existing on a “spectrum” of symptoms, which speaks to the wide range in clinical presentations (Hiremath et al., 2021; Kraguljac et al., 2021). Traumatic brain injury, as another example (6% of the studies here), varies in symptomatology, mechanism and location of injury, and severity and factors such as age at the time of injury and time postinjury. All of these independent factors may have profound consequences for neural systems and patient functioning (LaPlaca et al., 2020). To this point, few studies provided critical details regarding their samples to help address representativeness including education (35.6%), time since diagnosis (25.6%), age at diagnosis (10.8%), and race (5.6%) (see Supporting Information Table S2). The lack of clinical/demographic detail is of critical concern because even perfect prediction modeling by a classifier will leave open the question as to how the results will gen- eralize to other samples and undermines relevance for understanding clinical pathology. Modern data-sharing resources provide one opportunity to facilitate generalizable results by permitting clinical feature-dependent subgrouping. ENIGMA (Thompson et al., 2014, 2020), ADNI (Jack et al., 2008), ADHD200 (Di Martino et al., 2014), and OpenNeuro (Markiewicz et al., 2021) are all leading examples of data-sharing consortia that increase diversity of data collection sites, boost samples sizes, and enable representation clinical subgroups with respect to pathology chronicity and severity. While data sharing between sites poses challenges with respect to data harmonization (Radua et al., 2020), these factors (site/method) can be considered as features in prediction modeling. Brain Networks as Classifiers of Disease In network neuroscience, one of the biggest challenges is determining what the network should look like, including the number of nodes and how to define the links between them. This problem is no less evident in prediction modeling, where the machine is constrained by the complexity (or simplicity) of the representative neural network used for training. There has been much recent work and emerging consensus regarding best practices for fMRI data pre- processing (Esteban et al., 2019; Nichols et al., 2017; Zuo et al., 2019) and guidance for how networks might be reliably constructed and compared (Hallquist & Hillary, 2018; van den Heuvel et al., 2017; van Wijk et al., 2010). Even so, there remains a wide range of applications of network approaches and flexibility in workflows (i.e., investigator degrees of freedom; Gelman & Loken, 2014), which was evident in the current sampling of the literature. Just as one example, and consistent with the review by Hallquist and Hillary (2018), there was an enormous range in brain parcellation approaches with the number of nodes ranging from <10 to over 67k (see Table 2). The number of nodes in any network is a fundamental deter- minant for the downstream network characteristics such as path length, local clustering, degree, and even network strength (Bullmore & Bassett, 2011; Bullmore & Sporns, 2009; Network Neuroscience 40 Predictive modeling and network neuroscience van Wijk et al., 2010). Similarly, decisions about network sparsity and edge definition (e.g., correlation, partial correlation) hold crucial consequences for sensitivity to the clinical pathol- ogy (Figure 2, Decision 2). To address this issue investigators have actively modeled a range of network parameters (e.g., distinct brain parcellation approaches, distinct edge definitions) and data processing steps as part of prediction modeling to simultaneously advance the methods and isolate the role of investigator data processing decisions on model performance (Abraham et al., 2017; Badea et al., 2017; J. Chen et al., 2021; Rubbert et al., 2019). Examiners might be best advised to test distinct network thresholds and parcellations and share study-specific workflows (Botvinik-Nezer et al., 2020) with the goal of quantifying how choices made during data processing or network definition directly influence ML parameter- ization. Again, these decisions require explicit reporting so that consensus can be reached regarding best practices for using networks as inputs in prediction modeling (see Figure 2, Decision 2). Finally, studies of resting-state connectomics would likely benefit from recent machine learning advances in network representation learning (F. Chen et al., 2020; Hamilton et al., 2017) and predictive modeling from longitudinal data (Hsieh et al., 2020; Le & Honavar, 2020; Liang et al., 2020a, 2020b). There is also growing application of end-to-end deep learn- ing methods with a range of uses including data preprocessing (see Lu et al., 2021), dynamic connectivity modeling (see Fan et al., 2020), and structural connectomics (Sarwar et al., 2020). Transparency and Reporting Standards for Methods It was our original intention in this review to examine indicators of overfitting, feature engi- neering, hyperparameter determination, and other key decision points in ML. This goal was abandoned during our review because so few papers made transparent all of their steps during their training and classifier identification, decisions that should be consistently reported. The absence of these details for most studies is crucial and highlighted as a primary concern expressed in work by Hosseini et al. (2020), where a clear unintentional pathway to overfitting occurs in hyperparameter development, which permits pretraining exposure to data. Thorough reporting of the methodology surrounding the development of the ML process is key to understanding the value of the study and to allow meaningful replication analysis. For example, the reasons for algorithm selection and development, as well as the decisions made during training, can significantly impact the resultant performance and risk for overfitting (Cawley & Talbot, 2010; Hosseini et al., 2020). How features were selected should be shaped by the goals of the researcher and can shape the results of the experiment (Chandrashekar & Sahin, 2014), so the details of this step and its iterations require clear explanation in the method (Figure 2, Decisions 3 and 6). This will include, but is not necessarily limited to, the validation process, conditions for termination of the training loop, hyperparameter settings, any regularization or cost functions, and the iterative feature reduction techniques and param- eters (Figure 2, Decisions 4, 5, and 6, respectively). We propose that the reporting opportunities (R)presented in Figure 2 represent the standard minimum to assess how ML training was conducted and how feature inputs were selected. These questions, left unanswered, prevent an honest determination of overfitting risk and study general- izability in the absence of replication. There is real need in the literature reviewed here, and per- haps in others where prediction modeling is being used, to increase methodological transparency. To list the decisions made in the machine learning processing stream, and provide subsequent theoretical grounding for each, enables critical review of the algorithm. In addition, providing open access to the code before publication (Figure 2, Decision 3) not only enables greater repli- cability, but further allows for auditing of code, improving the chance of catching errors early. Network Neuroscience 41 Predictive modeling and network neuroscience Classifier Performance: The Pursuit of Classification Accuracy One of the most important issues to address in this literature is the emphasis on maximizing clas- sification accuracy. It must first be recognized that purely “black box” prediction has value, even where the goal is to maximize classification with little inference as to the reasons for algo- rithmic success (LeCun et al., 2015), and that there is often a natural trade-off between prediction and explanation (see Yarkoni & Westfall, 2017). The perspective in the current paper, however, is that to advance our understanding of brain disorders, neuroscientific theory must set the guiderails for predictive modeling and interpretation beyond prediction holds important value. To provide the logic for this perspective, we might imagine a scenario where clinical inves- tigators implement a highly complex deep learning algorithm to classify “responders” for a new drug for a brain disease. While there is immediate clinical value in a completely “black box” classifier that accurately separates responders from nonresponders, the algorithm affords no additional understanding of the disease. In this scenario there is no new information about why responders benefited from the treatment and, importantly, why nonresponders did not—information that can be used to improve both future prediction models and clinical inter- ventions. Ultimately, prediction modeling exists on an “explanation-prediction” continuum with some loss in transparency as one moves from less complex inferential models to more opaque models that are designed to optimize predictive performance (see Bzdok & Ioannidis, 2019). Investigators must strike a balance between ever-increasing algorithmic complexity and the need for understanding underlying mechanisms. Given this perspective, there are three issues to consider with respect to the focus on prediction accuracy in this literature. First, there was a nearly 15% drop-off in the performance from cross- validation test (i.e., internal) to lockbox performance. The reduced classification accuracy from training to test data set is unsurprising, but makes clear what has been known for some time: lock- box sets should be required in any published study in order to guard against overfitting and max- imize generalizability (Poldrack et al., 2020). We anticipate that overfitting is at least partially a consequence of the current overreliance on accuracy as the primary measure of an algorithm’s performance, but it also highlights an opportunity to improve machine learning practices. A second, and related, issue is that classification accuracy achieved in the absence of lock- box data (83% of the studies reviewed) presents the field with inflated validation results that become unrealistic benchmarks for others to match. In the current review, authors commonly compared accuracy of their work with known standards maintaining that it “outperformed” other algorithms. These comparisons have been formalized elsewhere in national and interna- tional competitions where investigators have been invited to apply ML to imaging data to accurately diagnose disorders such as schizophrenia (see Silva et al., 2014) and ADHD (see Brown et al., 2012). As outlined in a number of reviews, it is likely premature to believe that machine learning can currently fill any needed void as the primary diagnostic criterion for brain disorders (Mei et al., 2020; Pellegrini et al., 2018; Zhang-James et al., 2020). It is also unreasonable to assume that any single data classification algorithm will differentiate one group from another with near-perfect reliability, with the exception of severe clinical phenom- ena that are readily dissociated with standard clinical measurement (e.g., schizophrenia). Where classification can make a crucial impact, however, is by uncovering interactions between features that lead to novel pathways of discovery and intervention in the clinical neu- rosciences (more on this below). Finally, accuracy can be bolstered by additional metrics including sensitivity, likelihood ratio, log loss, and the AUROC are a good first step for their ability to expand on the base information given by accuracy (Poldrack et al., 2020). This added granularity uncovers Network Neuroscience 42 Predictive modeling and network neuroscience potential weaknesses of a given model, such as high degrees of type 1 or 2 errors, two issues that can be obscured in imbalanced datasets. Adding Context to Performance in Machine Learned Models In moving beyond classification performance, ML offers unparalleled opportunities to gain new insights into how neural systems adapt to brain injury and disease through use of techniques that highlight the contribution of the features to the algorithm’s performance and the significance of the model’s predictive ability. These measures add transparency to the training process (Sheikholeslami, 2019) and may help to uncover key underlying neural substrates previously unknown to the clinical field (Fawcett & Hoos, 2016). Furthermore, specific approaches, such as ablation studies, can even be used to identify new insights, essential aspects, and functional redundancies that contribute to the robustness of a given algorithm (Meyes et al., 2019). Some of the most commonly used machine learning algorithms in the field (SVM, RA) can integrate and appropriately weight the contributions of different features (Figure 3, Decision 7a, 7b). Despite this, use of these valuable tools remains limited, as highlighted in Table 4. Feature importance was the most commonly observed technique within the reviewed liter- ature, yet was implemented in less than half (47.2%) of studies. Similarly, measures that increase the robustness of the findings, such as determining consensus features (features that Consensus features: Features that are included a certain have been repeatedly selected across training iterations), were only adopted in a fifth (20.0%) threshold of training iterations, of examined studies. Both of these methods enable the reporting of the relative discriminative the more training iterations that the ability of specific features. Doing so allows clinical researchers to highlight patterns of specific algorithm selects the feature for, importance that can be linked to disease and injury phenotypes. the more likely that it is discriminative for classification. Ablation reports, included in roughly one-third (34.0%) of studies, examine the relative contribution of an individual feature (or set of features) on classification accuracy through selective removal and reanalysis. This metric is valuable for understanding when there are diminishing returns from overselecting features or for establishing a desired trade-off for model complexity against performance. Inclusion of an ablation report not only highlights the stabil- ity of algorithm performance, but also can establish baselines for necessary model complexity for reliable diagnosis of a given condition. Finally, by repeated randomization of class labels through a permutation analysis (used in 27.7% of studies), the risk of the final feature set being discriminative by chance alone can be assessed through a computed significance level. Such analyses measure the reliability of the feature set used for classification. There are tools that are available to investigators that permit stronger inferences about the reasons for classification success, though they were not commonly used in the papers reviewed here. A model in the absence of interpretation limits the ability for clinicians and researchers to build targeted treatments or identify risk factors that can be used to advance clinical. Understanding the contribution of specific features to classification success enables better stakeholder (e.g., neuropsychologists; neurologists) involvement concurrent with the development of models on the front end (Hedderich & Eickhoff, 2021) as well as explicit tech- niques to provide a clear explanation of the output of the classifier to the neuroscientist or physician at the back end (Heinrichs & Eickhoff, 2020). CONCLUSION Computational neuroscience holds the promise of fundamentally changing our understanding of brain disorders, and, with this promise, comes much deserved enthusiasm. However, the Network Neuroscience 43 Predictive modeling and network neuroscience use of ML techniques in network neuroscience reflects the growing pains observed when novel methods capture the excitement of a research community, something that the field of functional brain imaging has experienced before (Kosslyn et al., 1995; R. A. Poldrack et al., 2020; Vul et al., 2009). The ultimate goal for clinical neuroscience is to understand how the brain adapts to injury and disease, and ML can be leveraged to help uncover critical interac- tions between behavioral, genetic, blood-based biomarkers, imaging and other measurements of neural systems. While this literature review revealed examples where ML was being used in ways that may advance our understanding of clinical neuropathology, there is significant need for greater methodological transparency, better representation of heterogeneous clinical disor- ders in the training and testing datasets, and greater devotion to understanding mechanisms of neuropathology as opposed to binarized diagnostic success. In the case of the latter, there does appear to be a consequence for overemphasizing classification accuracy both in method and outcome. The generalizability of the results, replicability of the methods, and clinical value gained by the work can then be the guiding principles for our work. We have offered a few conservative recommendations in this review with the goal of continuing a dialog regarding how we can transition toward a more mature integration of ML into the neurosciences that accelerates our understanding of brain disorders and ultimately improves patient outcome. SUPPORTING INFORMATION Supporting information for this article is available at https://doi.org/10.1162/netn_a_00212. AUTHOR CONTRIBUTIONS Andrew Cwiek: Conceptualization; Data curation; Formal analysis; Investigation; Methodol- ogy; Writing – original draft; Writing – review & editing. Sarah Rajtmajer: Conceptualization; Investigation; Methodology; Resources; Writing – original draft; Writing – review & editing. Bradley Wyble: Conceptualization; Resources; Writing – original draft; Writing – review & editing. Vasant Honavar: Resources; Writing – original draft; Writing – review & editing. Emily Grossner: Formal analysis; Investigation; Writing – review & editing. Frank Hillary: Conceptu- alization; Data curation; Formal analysis; Investigation; Methodology; Project administration; Visualization; Writing – original draft; Writing – review & editing. REFERENCES Abadi, M., Barham, P., Chen, J., Chen, Z., Davis, A., Dean, J., ... Baldi, P., Baldi, P. P., Brunak, S., & Bach, F. (2001). Bioinformatics: Zheng, X. (2016). Tensorflow: A system for large-scale machine The machine learning approach. Cambridge, MA: MIT Press. learning. In 12th USENIX symposium on operating systems Baltrušaitis, T., Ahuja, C., & Morency, L. P. (2019). Multimodal design and implementation (OSDI 16) (pp. 265–283). machine learning: A survey and taxonomy. IEEE Transactions Abraham, A., Milham, M. P., Di Martino, A., Craddock, R. C., on Pattern Analysis and Machine Intelligence, 41(2), 423–443. Samaras, D., Thirion, B., & Varoquaux, G. (2017). Deriving https://doi.org/10.1109/TPAMI.2018.2798607 reproducible biomarkers from multi-site resting-state data: An Bassett, D. S., Cullen, K. E., Eickhoff, S. B., Farah, M. J., Goda, Y., Autism-based example. NeuroImage, 147, 736–745. https://doi Haggard, P., ... Ueda, H. R. (2020). Reflections on the past two .org/10.1016/j.neuroimage.2016.10.045, PubMed: 27865923 decades of neuroscience. Nature Reviews Neuroscience, 21(10), Abraham, A., Pedregosa, F., Eickenberg, M., Gervais, P., Mueller, A., 524–534. https://doi.org/10.1038/s41583-020-0363-6, PubMed: Kossaifi, J., ... Varoquaux, G. (2014). Machine learning for neu- 32879507 roimaging with scikit-learn. Frontiers in Neuroinformatics, 8, 14. Bishop, C. M. (2006). Pattern recognition and machine learning.CERN https://doi.org/10.3389/fninf.2014.00014, PubMed: 24600388 Document Server. Springer. https://cds.cern.ch/record/998831 Badea, L., Onu, M., Wu, T., Roceanu, A., & Bajenaru, O. (2017). Bonnelle, V., Ham, T. E., Leech, R., Kinnunen, K. M., Mehta, M. A., Exploring the reproducibility of functional connectivity alterations Greenwood, R. J., & Sharp, D. J. (2012). Salience network integ- in Parkinson’s disease. PLOS ONE, 12(11), e0188196. https://doi rity predicts default mode network function after traumatic brain .org/10.1371/journal.pone.0188196, PubMed: 29182621 injury. Proceedings of the National Academy of Sciences of the Network Neuroscience 44 Predictive modeling and network neuroscience United States of America, 109(12), 4690–4695. https://doi.org/10 patterns of task-defined brain networks allow individual predic- .1073/pnas.1113455109, PubMed: 22393019 tion of cognitive symptom dimension of schizophrenia and are Botvinik-Nezer, R., Holzmeister, F., Camerer, C. F., Dreber, A., linked to molecular architecture. Biological Psychiatry, 89(3), Huber, J., Johannesson, M., Kirchler, M., Iwanir, R., Mumford, 308–319. https://doi.org/10.1016/j.biopsych.2020.09.024, J. A., Adcock, R. A., Avesani, P., Baczkowski, B. M., Bajracharya, PubMed: 33357631 A., Bakst, L., Ball, S., Barilari, M., Bault, N., Beaton, D., Beitner, Chen, L., Chen, Y., Zheng, H., Zhang, B., Wang, F., Fang, J., Li, Y., J., … Schonberg, T. (2020). Variability in the analysis of a single Chen, Q., & Zhang, S. (2020). Changes in the topological orga- neuroimaging dataset by many teams. Nature, 582(7810), 84–88. nization of the default mode network in autism spectrum disor- https://doi.org/10.1038/s41586-020-2314-9,PubMed: der. Brain Imaging and Behavior. https://doi.org/10.1007/s11682 32483374 -020-00312-8, PubMed: 32737824 Bradski, G., & Kaehler, A. (2008). Learning OpenCV: Computer de Filippis, R., Carbone, E. A., Gaetano, R., Bruni, A., Pugliese, V., vision with the OpenCV Library.O’Reilly Media. Segura-Garcia, C., & De Fazio, P. (2019). Machine learning tech- Braun, U., Schaefer, A., Betzel, R. F., Tost, H., Meyer-Lindenberg, niques in a structural and functional MRI diagnostic approach in A., & Bassett, D. S. (2018). From maps to multi-dimensional net- schizophrenia: A systematic review. Neuropsychiatric Disease work mechanisms of mental disorders. Neuron, 97(1), 14–31. and Treatment, 15,1605–1627. https://doi.org/10.2147/NDT https://doi.org/10.1016/j.neuron.2017.11, PubMed: 29301099 .S202418, PubMed: 31354276 Brown, M. R. G, Sidhu, G. S., Greiner, R., Asgarian, N., Bastani, M., Deng, J., Dong, W., Socher, R., Li, L., Li, K., & Fei-Fei, L.. (2009). Silverstone, P. H., Greenshaw, A. J., & Dursun, S. M. (2012). ImageNet: A large-scale hierarchical image database. 2009 ADHD-200 Global Competition: Diagnosing ADHD using IEEE Conference on Computer Vision and Pattern Recognition, personal characteristic data can outperform resting state fMRI 248–255. https://doi.org/10.1109/CVPR.2009.5206848 measurements. Frontiers in Systems Neuroscience, 6. https://doi Dietterich, T. (1995). Overfitting and undercomputing in machine .org/10.3389/fnsys.2012.00069, PubMed: 23060754 learning. ACM Computing Surveys, 27(3), 326–327. https://doi Bullmore, E.T., & Bassett,D.S.(2011).Brain graphs:Graphical .org/10.1145/212094.212114 models of the human brain connectome. Annual Review of Clin- Di Martino, A.,Yan, C.G., Li,Q., Denio, E., Castellanos, F. X., ical Psychology, 7,113–140. https://doi.org/10.1146/annurev Alaerts, K., … Milham, M. P. (2014). The autism brain imaging -clinpsy-040510-143934, PubMed: 21128784 data exchange: Towards a large-scale evaluation of the intrinsic Bullmore, E. T., & Sporns, O. (2009). Complex brain networks: brain architecture in autism. Molecular Psychiatry, 19(6), 659–667. Graph theoretical analysis of structural and functional systems. https://doi.org/10.1038/mp.2013.78,PubMed: 23774715 Nature Reviews Neuroscience, 10(3), 186–198. https://doi.org Esteban, O., Markiewicz, C. J., Blair, R. W., Moodie, C. A., Isik, /10.1038/nrn2575, PubMed: 19190637 A. I.,Erramuzpe,A., … Gorgolewski, K. J. (2019). fMRIPrep: A Bzdok, D., & Ioannidis, J. P. A. (2019). Exploration, inference, and robust preprocessing pipeline for functional MRI. Nature prediction in neuroscience and biomedicine. Trends in Neurosci- Methods, 16(1), 111–116. https://doi.org/10.1038/s41592-018 ences, 42(4), 251–262. https://doi.org/10.1016/j.tins.2019.02 -0235-4, PubMed: 30532080 .001, PubMed: 30808574 Fan, L., Su, J., Qin, J., Hu, D., & Shen, H. (2020). A deep network Caeyenberghs, K., Verhelst, H., Clemente, A., & Wilson, P. H. model on dynamic functional connectivity with applications to (2017). Mapping the functional connectome in traumatic brain gender classification and intelligence prediction. Frontiers in injury: What can graph metrics tell us? NeuroImage, 160,113–123. Neuroscience, 14,881. https://doi.org/10.3389/fnins.2020 https://doi.org/10.1016/j.neuroimage.2016.12.003, PubMed: .00881, PubMed: 33013292 Fawcett, C., & Hoos, H. H. (2016). Analysing differences between Caruana, R., Lawrence, S., & Giles, C. (2000). Overfitting in neu- algorithm configurations through ablation. Journal of Heuristics, ral nets: Backpropagation, conjugate gradient, and early stop- 22(4), 431–458. https://doi.org/10.1007/s10732-014-9275-9 ping. Advances in Neural Information Processing Systems, 13, Forsyth, D., & Ponce, J. (2002). Computer vision: A modern 402–408. approach. Englewood Cliffs, NJ: Prentice-Hall. Cawley, G. C., & Talbot, N. L. C. (2010). On over-fitting in model Gao, J., Li, P., Chen, Z., & Zhang, J. (2020). A survey on deep selection and subsequent selection bias in performance evalua- learning for multimodal data fusion. Neural Computation, 32(5), tion. Journal of Machine Learning Research, 11(70), 2079–2107. 829–864. https://doi.org/10.1162/neco_a_01273, PubMed: Chandrashekar, G., & Sahin, F. (2014). A survey on feature selec- 32186998 tion methods. Computers & Electrical Engineering, 40(1), 16–28. Gelman, A., & Loken, E. (2014). The statistical crisis in science. https://doi.org/10.1016/j.compeleceng.2013.11.024 American Scientist, 102(6), 460–465. https://doi.org/10.1511 /2014.111.460 Chen, F., Wang, Y.-C., Wang, B., & Kuo, C.-C. J. (2020). Graph representation learning: A survey. APSIPA Transactions on Signal Gilbert, N., Bernier, R. A., Calhoun, V. D., Brenner, E., Grossner, E., and Information Processing, 9. https://doi.org/10.1017/ATSIP Rajtmajer, S. M., & Hillary, F. G. (2018). Diminished neural net- .2020.13 work dynamics after moderate and severe traumatic brain injury. PLoS One, 13(6), e0197419. https://doi.org/10.1371/journal Chen, J., Müller, V. I., Dukart, J., Hoffstaedter, F., Baker, J. T., .pone.0197419, PubMed: 29883447 Holmes, A. J., Vatansever, D., Nickl-Jockschat, T., Liu, X., Derntl, B., Kogler, L., Jardri, R., Gruber, O., Aleman, A., Sommer, I. E., Glerean, E., Pan, R. K., Salmi, J., Kujala, R., Lahnakoski, J. M., Roine, Eickhoff, S. B., & Patil, K. R. (2021). Intrinsic connectivity U., Nummenmaa, L., Leppämäki, S., Nieminen-von Wendt, T., Tani, Network Neuroscience 45 Predictive modeling and network neuroscience P., Saramäki, J., Sams, M., & Jääskeläinen, I. P. (2016). Reorganization C., Dale, A. M., Felmlee, J. P., Gunter, J. L., Hill, D. L. G, Killiany, of functionally connected brain subnetworks in high-functioning R., Schuff, N., Fox-Bosetti, S., Lin, C., Studholme, C., … Weiner, autism. Human Brain Mapping, 37(3), 1066–1079. https://doi.org M. W. (2008). The Alzheimer’s disease neuroimaging initiative /10.1002/hbm.23084,PubMed: 26686668 (ADNI): MRI methods. Journal of Magnetic Resonance Imaging, Goodfellow, I., Bengio, Y., & Courville, A. (2016). Deep learning 27(4), 685–691. https://doi.org/10.1002/jmri.21049,PubMed: (illustrated edition). Cambridge, MA: The MIT Press. 18302232 Guo, W., Wang, J., & Wang, S. (2019). Deep multimodal represen- Kosslyn, S. M., Thompson, W. L., Kim, I. J., & Alpert, N. M. (1995). tation learning: A survey. IEEE Access, 7, 63373–63394. https:// Topographical representations of mental images in primary visual doi.org/10.1109/ACCESS.2019.2916887 cortex. Nature, 378(6556), 496–498. https://doi.org/10.1038 Guyon, I., & Elisseeff, A. (2003). An introduction to variable and /378496a0, PubMed: 7477406 feature selection. The Journal of Machine Learning Research, 3, Kraguljac, N. V., McDonald, W. M., Widge, A. S., Rodriguez, C. I., 1157–1182. Tohen, M., & Nemeroff, C. B. (2021). Neuroimaging biomarkers Hallquist, M. N., & Hillary, F. G. (2018). Graph theory approaches in schizophrenia. The American Journal of Psychiatry. https://doi to functional network organization in brain disorders: A critique .org/10.1176/appi.ajp.2020.20030340, PubMed: 33397140 for a brave new small-world. Network Neuroscience, 3(1), 1–26. Langella, S., Sadiq, M. U., Mucha, P. J., Giovanello, K. S., Dayan, https://doi.org/10.1162/netn_a_00054, PubMed: 30793071 E., & Alzheimer’s Disease Neuroimaging Initiative. (2021). Lower Hamilton, W. L., Ying, R., & Leskovec, J. (2017). Representation functional hippocampal redundancy in mild cognitive impair- learning on graphs: Methods and applications. arXiv. https:// ment. Translational Psychiatry, 11(1), 61. https://doi.org/10.1038 arxiv.org/abs/1709.05584v3 /s41398-020-01166-w, PubMed: 33462184 Hastie, T., Tibshirani, R., & Friedman, J. (2009). The elements of LaPlaca, M. C., Huie, J. R., Alam, H. B., Bachstetter, A. D., Bayir, statistical learning: Data mining, inference, and prediction H., Bellgowan, P. S. F, Cummings, D., Dixon, C. E., Ferguson, (2nd ed.). Springer Science & Business Media. https://doi.org A. R., Ferland-Beckham, C., Floyd, C., Friess, S., Galanopoulou, /10.1007/978-0-387-84858-7 A., Hall, E. D., Harris, N. G., Hawkins, B. E., Hicks, R., Hulbert, Hawkins, D. M. (2004). The problem of overfitting. Journal of L. E., Johnson, V. E., … Zai, L. (2020). Preclinical common data Chemical Information and Computer Sciences, 44(1), 1–12. elements for traumatic brain injury research: Progress and use https://doi.org/10.1021/ci0342472, PubMed: 14741005 cases. Journal of Neurotrauma. https://doi.org/10.1089/neu Hedderich, D. M., & Eickhoff, S. B. (2021). Machine learning for .2020.7328, PubMed: 33297844 psychiatry: Getting doctors at the black box? Molecular Psychia- Larrañaga, P., Calvo, B., Santana, R., Bielza, C., Galdiano, J., Inza, try, 26(1), 23–25. https://doi.org/10.1038/s41380-020-00931-z, I., Lozano, J. A., Armañanzas, R., Santafé, G., Pérez, A., & Robles, PubMed: 33173196 V. (2006). Machine learning in bioinformatics. Briefings in Bioin- Hegarty, J. P., Ferguson, B. J., Zamzow, R. M., Rohowetz, L. J., Johnson, formatics, 7(1), 86–112. https://doi.org/10.1093/bib/bbk007, J. D., Christ, S. E., & Beversdorf, D. Q. (2017). Beta-adrenergic PubMed: 16761367 antagonism modulates functional connectivity in the default Le, T., & Honavar, V. (2020). Dynamical Gaussian process latent mode network of individuals with and without autism spectrum variable model for representation learning from longitudinal disorder. Brain Imaging and Behavior, 11(5), 1278–1289. https:// data. Proceedings of the 2020 ACM-IMS on Foundations of Data doi.org/10.1007/s11682-016-9604-8,PubMed: 27714553 Science Conference, 183–188. https://doi.org/10.1145/3412815 Heinrichs, B., & Eickhoff, S. B. (2020). Your evidence? Machine learning algorithms for medical diagnosis and prediction. Human LeCun, Y., Bengio, Y., & Hinton, G. (2015). Deep learning. Nature, Brain Mapping, 41(6), 1435–1444. https://doi.org/10.1002/hbm 521(7553), 436–444. https://doi.org/10.1038/nature14539, .24886, PubMed: 31804003 PubMed: 26017442 Hiremath, C. S., Sagar, K. J. V, Yamini, B. K., Girimaji, A. S., Kumar, Lefort-Besnard, J., Bassett, D. S., Smallwood, J., Margulies, D. S., R., Sravanti, S. L., Padmanabha, H., Vykunta Raju, K. N., Kishore, Derntl, B., Gruber, O., Aleman, A., Jardri, R., Varoquaux, G., M. T., Jacob, P., Saini, J., Bharath, R. D., Seshadri, S. P., & Kumar, Thirion, B., Eickhoff, S. B., & Bzdok, D. (2018). Different shades M. (2021). Emerging behavioral and neuroimaging biomarkers for of default mode disturbance in schizophrenia: Subnodal covari- early and accurate characterization of autism spectrum disorders: ance estimation in structure and function. Human Brain Map- A systematic review. Translational Psychiatry, 11(1), 42. https://doi ping, 39(2), 644–661. https://doi.org/10.1002/hbm.23870, .org/10.1038/s41398-020-01178-6, PubMed: 33441539 PubMed: 29105239 Hosseini, M., Powell, M., Collins, J., Callahan-Flintoft, C., Jones, W., Liang, J., Wu, Y., Xu, D., & Honavar, V. (2020a). Longitudinal deep Bowman, H., & Wyble, B. (2020). I tried a bunch of things: The kernel Gaussian process regression. ArXiv. https://arxiv.org/abs dangers of unexpected overfitting in classification of brain data. /2005.11770 Neuroscience and Biobehavioral Reviews, 119, 456–467. https:// Liang, J., Xu, D., Sun, Y., & Honavar, V. (2020b). LMLFM: longitu- doi.org/10.1016/j.neubiorev.2020.09.036, PubMed: 33035522 dinal multi-level factorization machine. Proceedings of the AAAI Hsieh, T.-Y., Wang, S., Sun, Y., & Honavar, V. (2020). Explainable Conference on Artificial Intelligence, 34(04), 4811–4818. https:// multivariate time series classification: A deep neural network doi.org/10.1609/aaai.v34i04.5916 which learns to attend to important variables as well as informa- Lu, H., Liu, S., Wei, H., Chen, C., & Geng, X. (2021). Deep tive time intervals. ArXiv. https://arxiv.org/abs/2011.11631 multi-kernel auto-encoder network for clustering brain functional Jack, C. R., Bernstein, M. A., Fox, N. C., Thompson, P., Alexander, connectivity data. Neural Networks, 135,148–157. https://doi.org G., Harvey, D., Borowski, B., Britson, P. J., Whitwell, J. L., Ward, /10.1016/j.neunet.2020.12.005,PubMed: 33388506 Network Neuroscience 46 Predictive modeling and network neuroscience Manning, C., Surdeanu, M., Bauer, J., Finkel, J., Bethard, S., & Standardization, 16(2), 119–127. https://doi.org/10.1016/0092 McClosky, D. (2014). The Stanford CoreNLP Natural Language -1157(88)90039-x, PubMed: 3131345 Processing Toolkit. Proceedings of 52nd Annual Meeting of the Radua, J., Vieta, E., Shinohara, R., Kochunov, P., Quidé, Y., Green, Association for Computational Linguistics: System Demonstra- M. J., Weickert, C. S., Weickert, T., Bruggemann, J., Kircher, T., tions,55–60. https://doi.org/10.3115/v1/P14-5010 Nenadić, I., Cairns, M. J., Seal, M., Schall, U., Henskens, F., Markiewicz, C. J.,Gorgolewski,K.J., Feingold, F.,Blair,R., Fullerton, J. M., Mowry, B., Pantelis, C., Lenroot, R., … van Erp, Halchenko, Y. O., Miller, E., … Poldrack, R. A. (2021). Open- T. (2020). Increased power by harmonizing structural MRI site Neuro: An open resource for sharing of neuroimaging data. differences with the ComBat batch adjustment method in bioRxiv. https://doi.org/10.1101/2021.06.28.450168 ENIGMA. NeuroImage, 218, 116956. https://doi.org/10.1016/j Mei, J., Desrosiers, C., & Frasnelli, J. (2020). Machine learning for .neuroimage.2020.116956, PubMed: 32470572 the diagnosis of Parkinson’s disease: A systematic review. ArXiv. Roelofs, R., Shankar, V., Recht, B., Fridovich-Keil, S., Hardt, M., https://arxiv.org/abs/2010.06101 Miller, J., & Schmidt, L. (2019). A meta-analysis of overfit- Meyes, R., Lu, M., de Puiseau, C. W., & Meisen, T. (2019). Ablation ting in machine learning. In Proceedings of the 33rd Inter- studies in artificial neural networks. ArXiv. https://arxiv.org/abs national Conference on Neural Information Processing Systems /1901.08644 (pp. 9179–9189). Mikolov, T., Chen, K., Corrado, G., & Dean, J. (2013). Efficient esti- Roy, A., Bernier, R. A., Wang, J., Benson, M., French, J. J., Good, mation of word representations in vector space. ArXiv. https:// D. C., & Hillary, F. G. (2017). The evolution of cost-efficiency arxiv.org/abs/1301.3781 in neural networks during recovery from traumatic brain injury. Min, S., Lee, B., & Yoon, S. (2017). Deep learning in bioinformatics. PLoS One, 12(4), e0170541. https://doi.org/10.1371/journal Briefings in Bioinformatics, 18(5), 851–869. https://doi.org/10 .pone.0170541, PubMed: 28422992 Rubbert, C., Mathys, C., Jockwitz, C., Hartmann, C. J., Eickhoff, S. B., .1093/bib/bbw068, PubMed: 27473064 Hoffstaedter, F., Caspers, S., Eickhoff, C. R., Sigl, B., Teichert, Mitchell, T. M. (1997). Machine learning (1st ed.). McGraw-Hill. N. A., Südmeyer, M., Turowski, B., Schnitzler, A., & Caspers, J. Ng, A. (1997). Preventing “overfitting” of cross-validation data. (2019). Machine-learning identifies Parkinson’s disease patients International Conference on Machine Learning, 97, 245–253. Nichols, T. E., Das, S., Eickhoff, S. B., Evans, A. C., Glatard, T., based on resting-state between-network functional connectivity. Hanke, M., Kriegeskorte, N., Milham, M. P., Poldrack, R. A., The British Journal of Radiology, 92(1101), 20180886. https:// Poline, J.-B., Proal, E., Thirion, B., Van Essen, D. C., White, T., & doi.org/10.1259/bjr.20180886, PubMed: 30994036 Yeo, B. T. T (2017). Best practices in data analysis and sharing in Salvatore, C., Cerasa, A., Battista, P., Gilardi, M. C., Quattrone, A., neuroimaging using MRI. Nature Neuroscience, 20(3), 299–303. & Castiglioni, I. (2015). Magnetic resonance imaging biomarkers https://doi.org/10.1038/nn.4500, PubMed: 28230846 for the early diagnosis of Alzheimer’s disease: A machine learn- Page, M. J., McKenzie, J. E., Bossuyt, P. M., Boutron, I., Hoffmann, T. C., ing approach. Frontiers in Neuroscience, 9. https://doi.org/10 Mulrow, C. D., Shamseer, L., Tetzlaff, J. M., Akl, E. A., Brennan, S. E., .3389/fnins.2015.00307, PubMed: 26388719 Chou, R., Glanville, J., Grimshaw, J. M., Hróbjartsson, A., Lalu, Sarwar, T., Seguin, C., Ramamohanarao, K., & Zalesky, A. (2020). M. M., Li, T., Loder, E. W., Mayo-Wilson, E., McDonald, S., … Towards deep learning for connectome mapping: A block decom- Moher, D. (2021). The PRISMA 2020 statement: An updated position framework. NeuroImage, 212, 116654. https://doi.org/10 guideline for reporting systematic reviews. The BMJ, 372, n71. .1016/j.neuroimage.2020.116654, PubMed: 32068163 https://doi.org/10.1136/bmj.n71, PubMed: 33782057 Schaffer, C. (1993). Overfitting avoidance as bias. Machine Learning, Parkes, L., Satterthwaite, T. D., & Bassett, D. S. (2020). Towards pre- 10(2), 153–178. https://doi.org/10.1007/BF00993504 cise resting-state fMRI biomarkers in psychiatry: Synthesizing Sheikholeslami, S. (2019). Ablation programming for machine developments in transdiagnostic research, dimensional models learning. DiVA. diva2:1349978. of psychopathology, and normative neurodevelopment. Current Silva, R. F., Castro, E., Gupta, C. N., Cetin, M., Arbabshirani, M., Opinion in Neurobiology, 65, 120–128. https://doi.org/10.1016 Potluru, V. K., Plis, S. M., & Calhoun, V. D. (2014). The tenth /j.conb.2020.10.016, PubMed: 33242721 annual MLSP competition: Schizophrenia classification chal- Pellegrini, E., Ballerini, L., Hernandez, M. del C. V., Chappell, lenge. 2014 IEEE International Workshop on Machine Learning F. M., González-Castro, V., Anblagan, D., Danso, S., Muñoz- for Signal Processing (MLSP),1–6. https://doi.org/10.1109/MLSP Maniega, S., Job, D., Pernet, C., Mair, G., MacGillivray, T. J., .2014.6958889 Trucco, E., & Wardlaw, J. M. (2018). Machine learning of neuro- Srivastava, N., Hinton, G., Krizhevsky, A., Sutskever, I., & Salakhutdinov, imaging for assisted diagnosis of cognitive impairment and R. (2014). Dropout: A simple way to prevent neural networks dementia: A systematic review. Alzheimer’s & Dementia: Diag- from overfitting. The Journal of Machine Learning Research, 15(1), nosis, Assessment & Disease Monitoring, 10, 519–535. https:// 1929–1958. doi.org/10.1016/j.dadm.2018.07.004, PubMed: 30364671 Steardo, L., Carbone, E. A., de Filippis, R., Pisanu, C., Segura-Garcia, Poldrack, R. A., Huckins, G., & Varoquaux, G. (2020). Establish- C., Squassina, A., De Fazio, P., & Steardo, L. (2020). Application ment of best practices for evidence for prediction: A review. of support vector machine on fMRI data as biomarkers in schizo- JAMA Psychiatry, 77(5), 534–540. https://doi.org/10.1001 phrenia diagnosis: A systematic review. Frontiers in Psychiatry, /jamapsychiatry.2019.3671, PubMed: 31774490 11, 588. https://doi.org/10.3389/fpsyt.2020.00588, PubMed: Progar, J. J., & May, J. C. (1988). The determination of low levels of 32670113 aluminum in antihemophilic factor (human) preparations by Thompson, P. M., Jahanshad, N., Ching, C. R. K, Salminen, L. E., flame atomic absorption spectrometry. Journal of Biological Thomopoulos, S. I., Bright, J., Baune, B. T., Bertolín, S., Bralten, Network Neuroscience 47 Predictive modeling and network neuroscience J., Bruin, W. B., Bülow, R., Chen, J., Chye, Y., Dannlowski, U., de using graph theory. PLoS One, 5(10), e13701. https://doi.org/10 Kovel, C. G. F, Donohoe, G., Eyler, L. T., Faraone, S. V., Favre, P., .1371/journal.pone.0013701, PubMed: 21060892 … ENIGMA Consortium. (2020). ENIGMA and global neurosci- Varoquaux, G., Raamana, P. R., Engemann, D. A., Hoyos-Idrobo, ence: A decade of large-scale studies of the brain in health and A., Schwartz, Y., & Thirion, B. (2017). Assessing and tuning brain disease across more than 40 countries. Translational Psychiatry, decoders: Cross-validation, caveats, and guidelines. Neuro- 10(1), 100. https://doi.org/10.1038/s41398-020-0705-1, Image, 145, 166–179. https://doi.org/10.1016/j.neuroimage PubMed: 32198361 .2016.10.038, PubMed: 27989847 Thompson, P. M., Stein, J. L., Medland, S. E., Hibar, D. P., Vasquez, Vu, M. A. T., Adali, T., Ba, D., Buzsáki, G., Carlson, D., Heller, K., ... A. A., Renteria, M. E., Toro, R., Jahanshad, N., Schumann, G., Dzirasa, K. (2018). A shared vision for machine learning in neu- Franke, B., Wright, M. J., Martin, N. G., Agartz, I., Alda, M., roscience. Journal of Neuroscience, 38(7), 1601–1607. https://doi Alhusaini, S., Almasy, L., Almeida, J., Alpert, K., Andreasen, .org/10.1523/JNEUROSCI.0508-17.2018, PubMed: 29374138 N. C., … Alzheimer’s Disease Neuroimaging Initiative, EPIGEN Vul, E., Harris, C., Winkielman, P., & Pashler, H. (2009). Puzzlingly Consortium, IMAGEN Consortium, Saguenay Youth Study (SYS) high correlations in fMRI studies of emotion, personality, and Group. (2014). The ENIGMA Consortium: Large-scale collabora- social cognition. Perspectives on Psychological Science, 4(3), tive analyses of neuroimaging and genetic data. Brain Imaging 274–290. https://doi.org/10.1111/j.1745-6924.2009.01125.x, and Behavior, 8(2), 153–182. https://doi.org/10.1007/s11682 PubMed: 26158964 -013-9269-5, PubMed: 24399358 Yarkoni, T., & Westfall, J. (2017). Choosing prediction over expla- Turian, J., Ratinov, L.-A., & Bengio, Y. (2010). Word representations: nation in psychology: Lessons from machine learning. Perspec- A simple and general method for semi-supervised learning. In tives on Psychological Science: A Journal of the Association for Proceedings of the 48th Annual Meeting of the Association for Psychological Science, 12(6), 1100–1122. https://doi.org/10 Computational Linguistics (pp. 384–394). .1177/1745691617693393, PubMed: 28841086 van den Heuvel, M. P., de Lange, S. C., Zalesky, A., Seguin, C., Yeo, Zhang-James, Y., Hoogman, M., Franke, B., & Faraone, S. V. (2020). B. T. T., & Schmidt, R. (2017). Proportional thresholding in Machine learning and MRI-based diagnostic models for ADHD: resting-state fMRI functional connectivity networks and conse- Arewethere yet? MedRxiv, 2020.10.20.20216390. https://doi quences for patient-control connectome studies: Issues and rec- .org/10.1101/2020.10.20.20216390 ommendations. NeuroImage, 152,437–449. https://doi.org/10 Zuo, X.-N., Biswal, B. B., & Poldrack, R. A. (2019). Editorial: Reli- .1016/j.neuroimage.2017.02.005, PubMed: 28167349 ability and reproducibility in functional connectomics. Frontiers van Wijk, B. C. M., Stam, C. J., & Daffertshofer, A. (2010). Compar- in Neuroscience, 13,117. https://doi.org/10.3389/fnins.2019 ing brain networks of different size and connectivity density .00117, PubMed: 30842722 Network Neuroscience 48 http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Network Neuroscience MIT Press

Feeding the machine: Challenges to reproducible predictive modeling in resting-state connectomics

Download PDF
20 pages

Loading next page...
 
/lp/mit-press/feeding-the-machine-challenges-to-reproducible-predictive-modeling-in-T30E3nlI1Q

References (216)

Publisher
MIT Press
Copyright
© 2021 Massachusetts Institute of Technology. Published under a Creative Commons Attribution 4.0 International (CC BY 4.0) license.
eISSN
2472-1751
DOI
10.1162/netn_a_00212
Publisher site
See Article on Publisher Site

Abstract

PERSPECTIVE Feeding the machine: Challenges to reproducible predictive modeling in resting-state connectomics 1,2 3,4 1 3,5 Andrew Cwiek , Sarah M. Rajtmajer , Bradley Wyble , Vasant Honavar , 1,2 1,2 Emily Grossner , and Frank G. Hillary Department of Psychology, Pennsylvania State University, University Park, PA, USA Social Life and Engineering Sciences Imaging Center, Pennsylvania State University, University Park, PA, USA College of Information Sciences and Technology, Pennsylvania State University, University Park, PA, USA Rock Ethics Institute, Pennsylvania State University, University Park, PA, USA Institute for Computational and Data Sciences, Pennsylvania State University, University Park, PA, USA Keywords: Machine learning, Classifiers, Predictive modeling, Brain networks, Clinical an open access journal neuroscience ABSTRACT In this critical review, we examine the application of predictive models, for example, classifiers, trained using machine learning (ML) to assist in interpretation of functional neuroimaging data. Our primary goal is to summarize how ML is being applied and critically assess common practices. Our review covers 250 studies published using ML and resting-state functional MRI (fMRI) to infer various dimensions of the human functional connectome. Results for holdout (“lockbox”) performance was, on average, ~13% less accurate than performance measured through cross-validation alone, highlighting the importance of lockbox Citation: Cwiek, A., Rajtmajer, S. M., data, which was included in only 16% of the studies. There was also a concerning lack of Wyble, B., Honavar, V., Grossner, E., & Hillary, F. G. (2022). Feeding the transparency across the key steps in training and evaluating predictive models. The summary machine: Challenges to reproducible predictive modeling in resting-state of this literature underscores the importance of the use of a lockbox and highlights several connectomics. Network Neuroscience, methodological pitfalls that can be addressed by the imaging community. We argue that, 6(1), 29–48. https://doi.org/10.1162/netn _a_00212 ideally, studies are motivated both by the reproducibility and generalizability of findings as well as the potential clinical significance of the insights. We offer recommendations for DOI: https://doi.org/10.1162/netn_a_00212 principled integration of machine learning into the clinical neurosciences with the goal of Supporting Information: advancing imaging biomarkers of brain disorders, understanding causative determinants for https://doi.org/10.1162/netn_a_00212 health risks, and parsing heterogeneous patient outcomes. Received: 12 February 2021 Accepted: 8 October 2021 Competing Interests: The authors have declared that no competing interests exist. BACKGROUND Corresponding Author: In settings where large amounts of well-characterized training data are available, modern Frank G. Hillary fhillary@psu.edu machine learning (ML) methods offer some of the most powerful approaches to discovering regularities and extracting useful knowledge from data (Bishop, 2006; Goodfellow et al., Handling Editor: Olaf Sporns 2016; Hastie et al., 2009; Mitchell, 1997). Of particular interest are algorithms that, given a data set of labeled samples, learn a predictive model, for example, a classifier, for labeling novel samples drawn from the same distribution as the training data. Programs for training Copyright: © 2021 such classifiers typically optimize a desired objective function on a given set of training sam- Massachusetts Institute of Technology Published under a Creative Commons ples. Advances in ML have revolutionized the design of systems for natural language process- Attribution 4.0 International ing (Manning et al., 2014; Mikolov et al., 2013; Turian et al., 2010), computer vision (Bradski (CC BY 4.0) license & Kaehler, 2008; Deng et al., 2009; Forsyth & Ponce, 2002), network analysis (Hamilton et al., The MIT Press Predictive modeling and network neuroscience Classifier: 2017), and bioinformatics (Baldi et al., 2001; Larrañaga et al., 2006; Min et al., 2017). A num- An algorithm designed to classify two ber of publicly available ML libraries (e.g., Scikit-learn, TensorFlow) can now be deployed, or more groups using a given set of permitting “off-the-shelf” application of these analyses for a number of data types including variables. behavioral, genetic, and imaging data (Abadi et al., 2016; Abraham et al., 2014). Training: In one sense, predictive models trained using ML are like traditional statistical models, for A generally iterative process wherein example, regression: there are covariates, an outcome, and a statistical function linking the an algorithm is refined to better covariates to the outcome. But where ML algorithms add value is in handling enormous num- classify a subject into their correct bers of features or predictors, heterogeneous data types (e.g., images, text, genomic sequences, group by tuning the parameters for molecular structures, networks, and longitudinal behavioral observations), and combining selection of important variables. them in complex, nonlinear ways to make accurate individualized prediction, that is, a clinical Features: diagnosis. This review examines the use of predictive models in ML and resting-state connec- The variables input into the algorithm tomics with focus on several particularly important issues, including “overfitting” and its for use in classification. related consequences, sample size and implications for modeling clinical heterogeneity, and methodological transparency. Overfitting: When the algorithm is too closely Prediction Modeling in the Neurosciences attuned to the data it was trained on, to the detriment of the algorithm’s There has been growing use of ML to determine if brain network metrics can serve as classi- generalizability to new samples. fiers of brain disorders with several high-profile reviews recently published (Bassettetal., 2020; Braun et al., 2018; Parkes et al., 2020; Vu et al., 2018). Many of the canonical networks identified in rsfMRI studies (e.g., default mode network) have been of critical focus in studies of large-scale network plasticity in a range of brain disorders including schizophrenia (de Filippis et al., 2019; Lefort-Besnard et al., 2018; Progar & May, 1988; Steardo et al., 2020), autism (L. Chen et al., 2020; Glerean et al., 2016; Hegarty et al., 2017), Alzheimer’s disease and related dementias (Langella et al., 2021; Pellegrini et al., 2018; Salvatore et al., 2015), and brain injury (Bonnelle et al., 2012; Caeyenberghs et al., 2017; Gilbert et al., 2018; Roy et al., 2017). Dimensionality: While the high dimensionality of functional imaging data—relationships between hundreds The number of features given to the or thousands of time series observations—may push the limits of traditional modeling, ML algorithm. approaches can capitalize on the complexity of multimodal datasets (Baltrušaitis et al., 2019; Gao et al., 2020; Guo et al., 2019) and provide opportunity to examine interactions among variables otherwise impossible to test. Therefore, there is evident potential for the application of ML to incorporate a wide array of data structures into prediction modeling including behavioral, brain imaging, physiological measurements, and genetic markers. Growing Pains in ML and Resting-State Connectomics Perhaps the most common methodological concern in applied ML is overfitting, or training an algorithm to predict with very high accuracy features within a single dataset at the expense of predicting a phenomenon more generally (Dietterich, 1995; Ng, 1997; Roelofs et al., 2019; Srivastava et al., 2014). Overfitting has profound implications for reproducibility, portability, and generalizability of findings. Importantly, the difficulty of preventing overfitting is underap- preciated, and even typical remedies, such as cross-validation, can allow for analysis hyper- Cross-validation: A process to limit overfitting through parameters to become tuned, or “overhyped,” to a specific set of data (Hosseini et al., 2020; repeated splitting of the data into Poldrack et al., 2020). These concerns underscore the need for greater transparency in model training and testing sets to prevent selection, enforcement of model parsimony, and rigorous testing and validation of trained models overfitting. on independent validation data, with attention to class imbalance in the data, relative costs of false positives versus false negatives, and the tradeoffs between them (Varoquaux et al., 2017). Related to overfitting are concerns about the size or heterogeneity in the training and test samples (Poldrack et al., 2020). When a sample is overly restrictive along dimensions that Network Neuroscience 30 Predictive modeling and network neuroscience influence outcome in neurological disorders (e.g., severity of disease, age of onset), it may reduce the study reproducibility and the ability to predict the phenomenon as it naturally occurs (Caruana et al., 2000; Hawkins, 2004; Schaffer, 1993; Srivastava et al., 2014). As an example, an investigator may have access to a large database of cases of individuals diagnosed with a neurological or psychiatric disorder that can be used for training and test datasets. Even with conservative training and only single exposure to the testing dataset (the gold standard), the result may not generalize if the sample is restricted in its range of characteristics with respect to demography, symptom severity, or disease/injury chronicity. Goals of this Review There is significant and justified enthusiasm for using ML approaches to advance our under- standing of brain disorders. With the ever-increasing application of ML in the study of resting- state connectomics, the importance of the implementation of and adherence to best practices is further underscored. Given this backdrop, we review 250 papers using ML for diagnosis or symptom profiling of brain disorders using resting-state fMRI methods, coding information regarding the methods used with particular focus on how algorithmic “success” was deter- mined, the use of a lockbox dataset (i.e., a data set that can be accessed only once at the Lockbox: A set of data intentionally set aside end of the analysis, also called a holdout set, a test set, or an external set), transparency in before training the algorithm and the approach, sample size and heterogeneity, and the types of conclusions drawn. We aim used exactly once after training to to provide a summary of the state-of-the-art in ML applications to one area of clinical neuro- test the generalizability of the result. science with the goal of identifying best practices and opportunities for methodological improvement. While we focus on resting-state fMRI connectomics here, the issues addressed likely have relevance for a wider range of ML applications in the neurosciences. Method: Literature Review We conducted a literature search using the following search terms in the PubMed database: (ML OR classifier OR supervised learn OR unsupervised learn OR SVM) AND (brain) AND (network OR graph OR connectivity) AND resting AND (imaging) AND (neurological OR clinical OR brain injury OR multiple sclerosis OR epilepsy OR stroke OR CVA OR aneurysm OR Parkinson’s OR MCI or Alzheimer’s OR dementia OR HIV OR SCI OR spinal cord OR autism OR ADHD OR intellectual disability OR Down syndrome OR Tourette) AND “humans”[MeSH Terms]. We did not bound the date range for our search, but we excluded non-English papers, review papers, and animal studies. We also excluded papers that were based on simulations or other nonhuman data. Our initial search returned 471 papers that were reviewed for inclu- sion. Two reviewers independently screened all of the papers returned from the above search at the title and abstract level for exclusionary criteria. By examining each paper title and abstract, papers were excluded based on the following categories: (1) examined structural brain imaging only (n = 98; 21%); (2) did not examine a clinical phenomenon (n = 59; 13%); (3) focused on automated tissue segmentation or lesion identification (n = 48, 10%); (4) was focused on algorithm or method development without clinical diagnostics (n = 41, 9%); (5) used other imaging approaches such as EEG/MEG (n = 33, 7%); (6) did not implement formal network analysis (n = 27, 6%); (7) was not an empirical study, including reviews and perspectives (n = 25, 5%); (8) did not use machine learning (broadly defined) or classification (n = 13, 3%); or (9) another reason consistent with the exclu- sionary criteria (n = 9, 2%). This resulted in exclusion of 353 papers, and for the remaining 118 papers (25%) the full paper was included in the final analysis. For the full-text review, two Network Neuroscience 31 Predictive modeling and network neuroscience reviewers were assigned to each specific section based on their respective specialties and completed a full analysis on multiple papers to identify any potential inconsistencies between the reviewers. Following this brief training for inter-rater consistency, the reviewers completed a full analysis of the papers independently. Based on feedback during the review process, we broadened our review to include terms sensitive to papers using deep learning approaches. A second identical keyword search to the above was conducted, while inserting the following terms to capture ML and deep learning approaches: “(deep learn* OR deep belief network OR multilayer perceptron OR autoencoder OR convolution neural network OR artificial neural network OR generative adversarial net- work OR machine learning OR ML OR classifier OR supervised learn OR unsupervised learn OR SVM) AND …”). The second search (April 2021) revealed 625 papers and based on abstract review (or full manuscript review if necessary), 405 papers were excluded based on the following categories and several for multiple reasons: (1) did not use machine learning (broadly defined) or classi- fication (179, 28.6%); (2) did not examine a clinical phenomenon (n = 90, 14.5%); (3) did not implement formal network analysis (n = 29, 4.6%); (4) used other imaging approaches such as EEG/MEG/PET (n = 28, 4.4%); (5) reviewed already existing literature, no new analysis (n = 24, 3.8%); (6) fMRI data were not included for prediction modeling (n = 22, 3.5%); (6) analysis included structural neuroimaging only (n = 12, 1.9%); (7) prospective study or proposal (n =6, .009%); (8) study not available in English (n = 3, .004%); (9) animal studies (n = 2, .003%); and (10) other reasons consistent with the exclusionary criteria (e.g., pilot studies, lesion segmen- tation studies, n = 11, .018%). This resulted in retention of 220 papers from our second search (n = 625). After eliminating redundancies with the outcome of the initial search (n = 471, n = 118 included), the final review included 250 unique papers for analysis. A flowchart for the literature review is provided in Figure 1. Data Coding To understand the methodological factors shaping machine learning use, the type of classifica- Permutation testing: tion algorithm utilized, subject population count, and the use of permutation testing with blind A method for testing the final feature analysis, as defined by Hosseini et al. (2020), were collected. Additionally, key information per- set against chance performance taining to the description of features input into the algorithm, the classifier design, and the per- through repeated randomization of formance reporting metrics chosen to measure chosen ML technique’s findings were collected. class labels (i.e., patient vs. healthy In addition to the year of publication, specific demographic factors of the participants used in control) and comparison of the distributed accuracy to the observed each paper were recorded. These factors include age, years of education, handedness, age of performance. diagnosis (where applicable), and socioeconomic status. Features used to train the algorithm were recorded including the use of network metrics, behavioral data, injury or disease charac- teristics, genetic information, blood biomarker information, medical history, and demographic factors. For network metrics specifically, information regarding the node definition and count, edge definition, and whole-brain versus subnetwork analysis were additionally recorded. Elements shaping the result reporting of the classifier, including the metrics chosen by the article, the type of cross-validation technique, ablation reporting, and use of a lockbox (i.e., a strictly observed separation between data used to train/optimize the analysis parameters and data used to assess generalizability; see Hosseini et al., 2020) were a primary focus of this review. Because classifier accuracy was a focus for our review, this was coded for all papers, and in the case of multiple analyses, the test with the highest performance at the most stringent Network Neuroscience 32 Predictive modeling and network neuroscience Figure 1. PRISMA flowchart of literature review. *An initial PubMed search was conducted, following valuable feedback, an updated search was conducted including articles up to the year 2021, and which included terms to broaden the search to include deep learning algorithms. For details, please see section Method: Literature Review. **Initial Review did not delineate removal at particular step; updated review includes a step-by-step workflow. ***220 from updated search + 30 nonduplicates from initial search. Modification of flowchart provided by Page et al. (2021). validation stage (cross-validation or lockbox) was selected for analysis. In cases where papers did not report total accuracy, but did report sensitivity and specificity, we calculated an accu- racy value based on sensitivity, specificity, and sample size (see Supporting Information For- mula: Formula S1). The presence of a lockbox was determined through keyword search of the full text for terms such as “holdout,”“external,”“test set,”“testing,”“withheld,” or “validation-set,” followed by a critical reading of the methodology. To qualify as a lockbox, the article had to (1) set aside a subset of data for the purpose of testing the algorithm performance following training, (2) make explicit that no part of the data in the lockbox was included at any point during algorithm development, and (3) not report multiple training/testing phases to arrive at the final lockbox performance. From the 250 papers, 44 (16.8%) included a test dataset, and of those, 32 included both lockbox and cross-validation performance reports. Network Neuroscience 33 Predictive modeling and network neuroscience Interpreting Model Performance Investigators have several tools at their disposal to better understand the impact of individual features on the final performance of the algorithm. While there are myriad ways in which spe- cific tools can be implemented, we searched for and coded the four most common methods observed in this literature set listed here in order of commonality: (1) feature importance, (2) Ablation analysis: permutation testing, (3) ablation analysis, and (4) consensus analysis. A measure of the contribution of Feature importance, or the discriminative power of an individual feature as assigned by the particular variables through manual trained algorithm, is an inherent element of many machine learning methodologies wherein removal or addition of specific features during training. features are ranked by their relative impact on the decision boundaries set by the algorithm. An article was coded as including feature importance if it included a report of some or all top- ranking features with some quantitative analysis of their relative contribution, such as (but not limited to) Gini index, Kendall’s tau values, or the correlation coefficient r. Permutation tests use thousands of randomized shufflings to simulate the distribution of pos- sible outcomes that a given comparison could have revealed if the independent variable was meaningless with respect to the analysis (i.e., the null hypothesis distribution). This technique can then measure the likelihood of an observed analysis outcome with an observed set of data or analysis outcome. Papers that run such analyses and report the likelihood of chance performance, generally in the form of p values, were coded as reporting this valuable analytical technique. An ablation analysis examines the performance of the algorithm when portions of the algo- rithm are removed in order to either improve performance (i.e., during training) or to determine which portions of the algorithm or dataset contribute to the algorithm’s accuracy. This is sim- ilar to feature selection in the context of neuroscience (Guyon & Elisseeff, 2003). For a paper to demonstrate an ablation report per our coding scheme, it must show the changes to perfor- mance in training, whether as a function of feature reduction or of iteration count. Consensus analysis is another common technique for analyzing relative importance of fea- tures by way of the ratio of times a feature is selected across the number of training/validation folds. Articles providing either a raw count or some other form of occurrence frequency for key features were coded as demonstrating a basic consensus analysis. RESULTS Representation of Clinical Disorders in Review The final review included 250 studies largely composed of case-control designs focused on prediction modeling of diagnostic accuracy. The studies ranged from traditional neurological diagnoses (Alzheimer’s disease, brain injury) to psychiatric disease (depression, anxiety), to neurodevelopmental diseases (schizophrenia, autism spectrum). A summary of the distinct clinical disorders represented in the review is provided in Supporting Information Table S1. The largest representation of studies examined Alzheimer’s disease or related dementias (n = 66, 26.4%), depression/anxiety (n = 40, 16.0%), schizophrenia spectrum disorder (n = 34, 13.6%), Autism spectrum disorder (n = 33, 13.2%), and brain injury (n = 15, 6.0%). For training, sample sizes ranged from 17 to 1,305 subjects for total samples and 8 to 653 for group-level data (case-control designs). For test datasets, the sample sizes for the total pop- ulation ranged from 8 to 477 subjects and group-level data ranged from 1 to 185. See Table 1 for breakdown of training and test dataset sample sizes based on population and subgroup. These sample sizes are consistent with those observed elsewhere (Poldrack et al., 2020), and we anticipate that the large majority of studies present in this review were underpowered Network Neuroscience 34 Predictive modeling and network neuroscience Table 1. Sample sizes for population and subgroups in training and test datasets Training set (n = 250) Test set (n = 44) Sample Total Subgroup Total Subgroup Range 17–1305 8–653 8–477 1–185 Mean 126.7 50.0 96.6 38.1 Median 77 29 39 20 Studies with n ≤ 50 80 (32.0%) 192 (76.8%) 23 (52.3%) 35 (79.6%) Studies with n ≤ 30 24 (9.6%) 136 (54.4%) 14 (31.8%) 28 (63.6) Studies with n ≤ 20 3 (1.2%) 82 (32.8%) 8 (18.2%) 22 (50.0%) for reliable prediction modeling, resulting in low confidence in the portability of the reported algorithm and reproducibility of the finding in other samples. Network Characteristics Consistent with the inclusionary criteria, 100% of the studies used at least one network metric as input during classifier identification. Table 2 provides descriptive data for the types of network studies included and the characteristics of the networks analyzed. A majority of the studies used whole-brain network information as features (73%). Similar to other exami- nations of the use of network neuroscience to examine clinical disorders, there was a wide range of brain parcellation values, resulting in graphs of widely varying sizes and complexities (Hallquist & Hillary, 2018). Sample Characteristics Sample characteristics including demographics and common clinical indicators were exam- ined. While age of the sample was commonly reported, only 25.6% of studies included a mea- sure of time since diagnosis, fewer still reported age of diagnosis (10.8%), and few included demographic factors such as race (5.6%). Several studies lacked a “healthy control” group. In these cases, the studies either compared the same sample at two timepoints (1) or classified Table 2. Network data: Characteristics of functional brain imaging network analysis including in prediction modeling Range Median Mean (SD) Mode Network Nodes (parcellation) <10 to 67,955 90 483.9 (6,654.5) 90 n = 221* Correlation Partial Causal (e.g., Pearson’s r) correlation Multiple modeling Other Edge Definition n = 247* 67.9% 3.2% 6.1% 3.6% 18.3% Whole brain Modules/subnetworks Nodes/seed-based Unclear Scope of study n = 250 73.1% 19.0% 7.9% 3% * Note: All studies included defined nodes, but in some cases the exact number of nodes was unclear with respect to ML training (n = 30). Similarly, all studies examined connectivity between brain regions, but for a small number of studies there was no clear edge definition (n = 3). Network Neuroscience 35 Predictive modeling and network neuroscience against two separate clinical groups (5). See Supporting Information Table S2 for details regarding reported demography. Classifier Information Critical to our goal was to assess the classifiers used in this literature, the most common input features, and how the classifiers were evaluated. Table 3 provides summary information for characteristics of classifiers used across the 250 studies. Support vector machines were the most prevalent ML algorithm selected, appearing as at least one of the utilized classifiers in 68.4% of papers. The three next most common techniques used were linear discriminant analysis (8.8%), regression classification (8.8%), random forest (8.0%), and deep learning approaches (8%). In the papers reviewed, 18.8% implemented multiple ML algorithms. The feature types chosen to be used for a classifier, while not inherent to the quality of the ML algorithm, do speak to the nuances of the aims of the collected studies. While every study collected some form of injury or disease characteristic (at the very least for the purpose of identifying patient and healthy control groups), roughly 8% of studies included some form of these metrics as features to include in the classifier, and even fewer included demographic information (7%) and/or behavioral or cognitive data (4%) as features for training. Medication history, current medications, or other clinical metrics were rarely included as features (<1%). Only one study utilized blood biomarkers, and none included genetic factors in addition to network metrics, revealing an as-of-yet untouched potential for more nuanced feature-pattern discovery. Accuracy was the primary performance metric, with roughly 86.8% of papers reporting accuracy as the baseline measure of performance. More than two-thirds of studies included sensitivity and specificity, two metrics vital for identifying bias in classification, especially in the case of sample imbalance, whereas only 40.0% of studies included a full area under the receiver operating characteristics curve (AUROC) report; 12.0% of studies included predictive power, and 12.0% of studies included some form of regression analysis in addition to the Table 3. Classifier types, inputs, and metrics for evaluation during classification Deep Classifier SVM RF KNN LOG_R LDA learning Multiple Other Frequency* 171 (68.4%) 20 (8.0%) 17 (6.8%) 22 (8.8%) 22 (8.8%) 20 (8.0%) 46 52 (20.8%) (18.0%) Brain Injury/ Behavior/ Genes/ Inputs into network disease cognitive Medical blood classifier metrics actor Demographic data Hx Meds biomarkers Other Frequency 100% 13.5% 10.1% 5.9% 2.5% 1.7% 0% 1.6% Metric for AUC Predictive Regression evaluation Accuracy Sensitivity Specificity (AUROC) power outputs Other (e.g., F1) Frequency 87% 70.4% 69% 40% 12% 12% 20% Note: SVM, support vector machine; RF, random forest; KNN, k nearest-neighbor; LOG_R, logistic regression; LDA, linear discriminant analysis. *Total >100%, including studies with more than one classification approach. Network Neuroscience 36 Predictive modeling and network neuroscience Table 4. Validation measures Validation procedures Yes No Unclear Cross-validation 94.1% 4.2% 1.7% Lockbox 20.3% 79.7% 0.0% If lockbox, compared 70.8% 12.5% 16.7% once (n = 24) classification outputs. Finally, 20.8% of studies utilized some other form of metric performance reporting, such as F1 scores; all such measures fitting the “other” category were utilized in less than 5% of papers. Validation Approaches Most studies utilized some form of cross-validation, including leave-one-out cross-validation (LOOCV) (58.8%), k-fold (35.6%), nested approaches (11.2%), and multiple approaches (9.2%). Of note, 12 (4.8%) of the studies did not report any cross-validation use. In these cases, the authors either provided no alternative validation method (n = 8) or used a lockbox but no cross-validation (n = 4). The key diagnostic for overfitting, use of a lockbox, was only utilized in 16.8% of studies (Table 4). Of the studies using a lockbox, 81% (34/44) made clear that iterative training never permitted access to the test (lockbox) data, and 73.8% (31/44) reported accuracy results for both the training and lockbox data. Interpreting Model Performance Feature importance measures were the most common metric included, with nearly half of all studies including some level of quantitative analysis (47.2%). The other three common tech- niques for model interpretation were observed at a rate ranging between 1-in-3 to 1-in-5 papers. Permutation testing was included in 34.0% of all studies. Ablation reports were included in 27.7%, and consensus analyses were utilized in 20.0% of all studies (see Table 5). It was rare for examiners to include some form of all four approaches described here (2.8%), but about one-third of papers integrated two to three techniques (35.2%), more than a third integrated at least one method (38.4%), and finally one-fifth of papers did not conduct an analysis of feature importance (22.8%). Table 5. Common techniques for enhancing model interpretation Model interpretation techniques Yes No Feature importance 47.2% 52.8% Permutation testing 34.0% 66.0% Ablation analysis 27.7% 72.3% Consensus features 20.0% 80.0% Note: >100% due to multiple approaches used in some studies. Network Neuroscience 37 Predictive modeling and network neuroscience Figure 2. A histogram of accuracy scores for n = 250 studies reviewed reveals distinct distributions and median scores (organized in text boxes by color) for classification accuracy based on results using no validation, cross-validation, and external validation (i.e., lockbox). Classifier Performance Measuring ML performance with no form of cross-validation or lockbox validation produced a median accuracy of 97.1%. ML application using a cross-validation produced a median clas- sification accuracy of 86.8%. When classification was performed on lockbox data, the median classification accuracy dropped to 73.9%. The distribution for accuracy values across these distinct cross-validation approaches is reported in Figure 2. DISCUSSION While our review confirms the exciting promise of ML approaches in the network neurosciences to advance overall understanding of brain disorders, there also appears to be room for method- ological growth. We first make several observations regarding clinical sampling and how net- work neuroscience has been implemented in this literature as inputs for predictive modeling. We then focus the remainder of the discussion on critical issues that, if addressed, can bring greater precision to the use of ML in the neurosciences and ideally accelerate our understanding of the pathophysiology of brain disorders. In the following we highlight several issues in order to foster discussion in the literature: (1) need for uniformity in the creation of neural networks for predic- tion, (2) issues of sample size and heterogeneity, (3) need for greater transparency of methods and reporting standards, (4) the focus on classification accuracy at the expense of other informa- tion, and (5) explainability and feature importance. We outline these concerns and link them to eight decision points in the typical ML processing stream outlined in Figure 3, which serves as a roadmap for key considerations and reporting opportunities at each step of the training process with the goal of improving the interpretability, reproducibility, and clinical utility. Network Neuroscience 38 Predictive modeling and network neuroscience Figure 3. Illustration of distinct decision points in the typical ML pipeline in the papers included in this review. We identify eight distinct decision points where there are opportunities to report (R) information to maximize transparency. R1a: Justify classifier model choice from previous literature, limitations of data, and clinical goals of study. R1b: Explain how data were split between training and test sets (i.e., lockbox), including sample sizes and any matching of demographics or disease variables. R2: Make clear decisions about how the network was created, including edge definition and brain parcellation. R3: Make explicit the specifics of the model (e.g., parameter settings, kernel functions). Make clear which features (e.g., network metrics, clinical variables) are included in the model. R4: Report cross-validation method selection and implementation; justify use in context of sample size and potential risk of performance overestimation. R5: Explain the conditions necessary to terminate algorithm training, such as target performance or minimal feature count. R6:Make explicit the hyperparameter settings and any manual tuning of parameters between training itera- Feature weight: tions. R7a: Report training set results, including model performance, feature weights, and feature The discriminative ability of a given counts across training iterations. R7b: Explicitly state that preprocessing is unchanged from the final feature as measured and quantified algorithm derived from training and that during training there was no access to the lockbox; provide through various methodologies. the final averaged cross-validation performance and feature importance for the test set. R8: Provide clear interpretation and explainability for the model by highlighting any key findings in context of potential clinical utility (i.e., relevant regions of interest’s connectivity patterns). Network Neuroscience 39 Predictive modeling and network neuroscience Sample Sizes and Clinical Heterogeneity Roughly one-third of the studies sampled in this review had no more than 50 subjects in their total sample size for use of training and internal validation of their results. Furthermore, half of all lockbox sets examined had subgroup sample sizes of 20 or less. Thus, roughly half of the studies reviewed were likely underpowered to capture the stage, severity, and symptom con- stellation evident in heterogeneous neurological and neuropsychiatric disorders. Moreover, small samples likely contributed to the use of LOOCV (58.8%) instead of k-fold (35.6%), which may be more representative of the dataset (Poldrack et al., 2020). Clinical characteristics of the participants (representativeness) that comprise a sample may be just as vital as the sample size. Most neurological disorders maintain heterogeneous pre- sentations. For example, over a quarter of the studies focused on either schizophrenia or autism, both understood as existing on a “spectrum” of symptoms, which speaks to the wide range in clinical presentations (Hiremath et al., 2021; Kraguljac et al., 2021). Traumatic brain injury, as another example (6% of the studies here), varies in symptomatology, mechanism and location of injury, and severity and factors such as age at the time of injury and time postinjury. All of these independent factors may have profound consequences for neural systems and patient functioning (LaPlaca et al., 2020). To this point, few studies provided critical details regarding their samples to help address representativeness including education (35.6%), time since diagnosis (25.6%), age at diagnosis (10.8%), and race (5.6%) (see Supporting Information Table S2). The lack of clinical/demographic detail is of critical concern because even perfect prediction modeling by a classifier will leave open the question as to how the results will gen- eralize to other samples and undermines relevance for understanding clinical pathology. Modern data-sharing resources provide one opportunity to facilitate generalizable results by permitting clinical feature-dependent subgrouping. ENIGMA (Thompson et al., 2014, 2020), ADNI (Jack et al., 2008), ADHD200 (Di Martino et al., 2014), and OpenNeuro (Markiewicz et al., 2021) are all leading examples of data-sharing consortia that increase diversity of data collection sites, boost samples sizes, and enable representation clinical subgroups with respect to pathology chronicity and severity. While data sharing between sites poses challenges with respect to data harmonization (Radua et al., 2020), these factors (site/method) can be considered as features in prediction modeling. Brain Networks as Classifiers of Disease In network neuroscience, one of the biggest challenges is determining what the network should look like, including the number of nodes and how to define the links between them. This problem is no less evident in prediction modeling, where the machine is constrained by the complexity (or simplicity) of the representative neural network used for training. There has been much recent work and emerging consensus regarding best practices for fMRI data pre- processing (Esteban et al., 2019; Nichols et al., 2017; Zuo et al., 2019) and guidance for how networks might be reliably constructed and compared (Hallquist & Hillary, 2018; van den Heuvel et al., 2017; van Wijk et al., 2010). Even so, there remains a wide range of applications of network approaches and flexibility in workflows (i.e., investigator degrees of freedom; Gelman & Loken, 2014), which was evident in the current sampling of the literature. Just as one example, and consistent with the review by Hallquist and Hillary (2018), there was an enormous range in brain parcellation approaches with the number of nodes ranging from <10 to over 67k (see Table 2). The number of nodes in any network is a fundamental deter- minant for the downstream network characteristics such as path length, local clustering, degree, and even network strength (Bullmore & Bassett, 2011; Bullmore & Sporns, 2009; Network Neuroscience 40 Predictive modeling and network neuroscience van Wijk et al., 2010). Similarly, decisions about network sparsity and edge definition (e.g., correlation, partial correlation) hold crucial consequences for sensitivity to the clinical pathol- ogy (Figure 2, Decision 2). To address this issue investigators have actively modeled a range of network parameters (e.g., distinct brain parcellation approaches, distinct edge definitions) and data processing steps as part of prediction modeling to simultaneously advance the methods and isolate the role of investigator data processing decisions on model performance (Abraham et al., 2017; Badea et al., 2017; J. Chen et al., 2021; Rubbert et al., 2019). Examiners might be best advised to test distinct network thresholds and parcellations and share study-specific workflows (Botvinik-Nezer et al., 2020) with the goal of quantifying how choices made during data processing or network definition directly influence ML parameter- ization. Again, these decisions require explicit reporting so that consensus can be reached regarding best practices for using networks as inputs in prediction modeling (see Figure 2, Decision 2). Finally, studies of resting-state connectomics would likely benefit from recent machine learning advances in network representation learning (F. Chen et al., 2020; Hamilton et al., 2017) and predictive modeling from longitudinal data (Hsieh et al., 2020; Le & Honavar, 2020; Liang et al., 2020a, 2020b). There is also growing application of end-to-end deep learn- ing methods with a range of uses including data preprocessing (see Lu et al., 2021), dynamic connectivity modeling (see Fan et al., 2020), and structural connectomics (Sarwar et al., 2020). Transparency and Reporting Standards for Methods It was our original intention in this review to examine indicators of overfitting, feature engi- neering, hyperparameter determination, and other key decision points in ML. This goal was abandoned during our review because so few papers made transparent all of their steps during their training and classifier identification, decisions that should be consistently reported. The absence of these details for most studies is crucial and highlighted as a primary concern expressed in work by Hosseini et al. (2020), where a clear unintentional pathway to overfitting occurs in hyperparameter development, which permits pretraining exposure to data. Thorough reporting of the methodology surrounding the development of the ML process is key to understanding the value of the study and to allow meaningful replication analysis. For example, the reasons for algorithm selection and development, as well as the decisions made during training, can significantly impact the resultant performance and risk for overfitting (Cawley & Talbot, 2010; Hosseini et al., 2020). How features were selected should be shaped by the goals of the researcher and can shape the results of the experiment (Chandrashekar & Sahin, 2014), so the details of this step and its iterations require clear explanation in the method (Figure 2, Decisions 3 and 6). This will include, but is not necessarily limited to, the validation process, conditions for termination of the training loop, hyperparameter settings, any regularization or cost functions, and the iterative feature reduction techniques and param- eters (Figure 2, Decisions 4, 5, and 6, respectively). We propose that the reporting opportunities (R)presented in Figure 2 represent the standard minimum to assess how ML training was conducted and how feature inputs were selected. These questions, left unanswered, prevent an honest determination of overfitting risk and study general- izability in the absence of replication. There is real need in the literature reviewed here, and per- haps in others where prediction modeling is being used, to increase methodological transparency. To list the decisions made in the machine learning processing stream, and provide subsequent theoretical grounding for each, enables critical review of the algorithm. In addition, providing open access to the code before publication (Figure 2, Decision 3) not only enables greater repli- cability, but further allows for auditing of code, improving the chance of catching errors early. Network Neuroscience 41 Predictive modeling and network neuroscience Classifier Performance: The Pursuit of Classification Accuracy One of the most important issues to address in this literature is the emphasis on maximizing clas- sification accuracy. It must first be recognized that purely “black box” prediction has value, even where the goal is to maximize classification with little inference as to the reasons for algo- rithmic success (LeCun et al., 2015), and that there is often a natural trade-off between prediction and explanation (see Yarkoni & Westfall, 2017). The perspective in the current paper, however, is that to advance our understanding of brain disorders, neuroscientific theory must set the guiderails for predictive modeling and interpretation beyond prediction holds important value. To provide the logic for this perspective, we might imagine a scenario where clinical inves- tigators implement a highly complex deep learning algorithm to classify “responders” for a new drug for a brain disease. While there is immediate clinical value in a completely “black box” classifier that accurately separates responders from nonresponders, the algorithm affords no additional understanding of the disease. In this scenario there is no new information about why responders benefited from the treatment and, importantly, why nonresponders did not—information that can be used to improve both future prediction models and clinical inter- ventions. Ultimately, prediction modeling exists on an “explanation-prediction” continuum with some loss in transparency as one moves from less complex inferential models to more opaque models that are designed to optimize predictive performance (see Bzdok & Ioannidis, 2019). Investigators must strike a balance between ever-increasing algorithmic complexity and the need for understanding underlying mechanisms. Given this perspective, there are three issues to consider with respect to the focus on prediction accuracy in this literature. First, there was a nearly 15% drop-off in the performance from cross- validation test (i.e., internal) to lockbox performance. The reduced classification accuracy from training to test data set is unsurprising, but makes clear what has been known for some time: lock- box sets should be required in any published study in order to guard against overfitting and max- imize generalizability (Poldrack et al., 2020). We anticipate that overfitting is at least partially a consequence of the current overreliance on accuracy as the primary measure of an algorithm’s performance, but it also highlights an opportunity to improve machine learning practices. A second, and related, issue is that classification accuracy achieved in the absence of lock- box data (83% of the studies reviewed) presents the field with inflated validation results that become unrealistic benchmarks for others to match. In the current review, authors commonly compared accuracy of their work with known standards maintaining that it “outperformed” other algorithms. These comparisons have been formalized elsewhere in national and interna- tional competitions where investigators have been invited to apply ML to imaging data to accurately diagnose disorders such as schizophrenia (see Silva et al., 2014) and ADHD (see Brown et al., 2012). As outlined in a number of reviews, it is likely premature to believe that machine learning can currently fill any needed void as the primary diagnostic criterion for brain disorders (Mei et al., 2020; Pellegrini et al., 2018; Zhang-James et al., 2020). It is also unreasonable to assume that any single data classification algorithm will differentiate one group from another with near-perfect reliability, with the exception of severe clinical phenom- ena that are readily dissociated with standard clinical measurement (e.g., schizophrenia). Where classification can make a crucial impact, however, is by uncovering interactions between features that lead to novel pathways of discovery and intervention in the clinical neu- rosciences (more on this below). Finally, accuracy can be bolstered by additional metrics including sensitivity, likelihood ratio, log loss, and the AUROC are a good first step for their ability to expand on the base information given by accuracy (Poldrack et al., 2020). This added granularity uncovers Network Neuroscience 42 Predictive modeling and network neuroscience potential weaknesses of a given model, such as high degrees of type 1 or 2 errors, two issues that can be obscured in imbalanced datasets. Adding Context to Performance in Machine Learned Models In moving beyond classification performance, ML offers unparalleled opportunities to gain new insights into how neural systems adapt to brain injury and disease through use of techniques that highlight the contribution of the features to the algorithm’s performance and the significance of the model’s predictive ability. These measures add transparency to the training process (Sheikholeslami, 2019) and may help to uncover key underlying neural substrates previously unknown to the clinical field (Fawcett & Hoos, 2016). Furthermore, specific approaches, such as ablation studies, can even be used to identify new insights, essential aspects, and functional redundancies that contribute to the robustness of a given algorithm (Meyes et al., 2019). Some of the most commonly used machine learning algorithms in the field (SVM, RA) can integrate and appropriately weight the contributions of different features (Figure 3, Decision 7a, 7b). Despite this, use of these valuable tools remains limited, as highlighted in Table 4. Feature importance was the most commonly observed technique within the reviewed liter- ature, yet was implemented in less than half (47.2%) of studies. Similarly, measures that increase the robustness of the findings, such as determining consensus features (features that Consensus features: Features that are included a certain have been repeatedly selected across training iterations), were only adopted in a fifth (20.0%) threshold of training iterations, of examined studies. Both of these methods enable the reporting of the relative discriminative the more training iterations that the ability of specific features. Doing so allows clinical researchers to highlight patterns of specific algorithm selects the feature for, importance that can be linked to disease and injury phenotypes. the more likely that it is discriminative for classification. Ablation reports, included in roughly one-third (34.0%) of studies, examine the relative contribution of an individual feature (or set of features) on classification accuracy through selective removal and reanalysis. This metric is valuable for understanding when there are diminishing returns from overselecting features or for establishing a desired trade-off for model complexity against performance. Inclusion of an ablation report not only highlights the stabil- ity of algorithm performance, but also can establish baselines for necessary model complexity for reliable diagnosis of a given condition. Finally, by repeated randomization of class labels through a permutation analysis (used in 27.7% of studies), the risk of the final feature set being discriminative by chance alone can be assessed through a computed significance level. Such analyses measure the reliability of the feature set used for classification. There are tools that are available to investigators that permit stronger inferences about the reasons for classification success, though they were not commonly used in the papers reviewed here. A model in the absence of interpretation limits the ability for clinicians and researchers to build targeted treatments or identify risk factors that can be used to advance clinical. Understanding the contribution of specific features to classification success enables better stakeholder (e.g., neuropsychologists; neurologists) involvement concurrent with the development of models on the front end (Hedderich & Eickhoff, 2021) as well as explicit tech- niques to provide a clear explanation of the output of the classifier to the neuroscientist or physician at the back end (Heinrichs & Eickhoff, 2020). CONCLUSION Computational neuroscience holds the promise of fundamentally changing our understanding of brain disorders, and, with this promise, comes much deserved enthusiasm. However, the Network Neuroscience 43 Predictive modeling and network neuroscience use of ML techniques in network neuroscience reflects the growing pains observed when novel methods capture the excitement of a research community, something that the field of functional brain imaging has experienced before (Kosslyn et al., 1995; R. A. Poldrack et al., 2020; Vul et al., 2009). The ultimate goal for clinical neuroscience is to understand how the brain adapts to injury and disease, and ML can be leveraged to help uncover critical interac- tions between behavioral, genetic, blood-based biomarkers, imaging and other measurements of neural systems. While this literature review revealed examples where ML was being used in ways that may advance our understanding of clinical neuropathology, there is significant need for greater methodological transparency, better representation of heterogeneous clinical disor- ders in the training and testing datasets, and greater devotion to understanding mechanisms of neuropathology as opposed to binarized diagnostic success. In the case of the latter, there does appear to be a consequence for overemphasizing classification accuracy both in method and outcome. The generalizability of the results, replicability of the methods, and clinical value gained by the work can then be the guiding principles for our work. We have offered a few conservative recommendations in this review with the goal of continuing a dialog regarding how we can transition toward a more mature integration of ML into the neurosciences that accelerates our understanding of brain disorders and ultimately improves patient outcome. SUPPORTING INFORMATION Supporting information for this article is available at https://doi.org/10.1162/netn_a_00212. AUTHOR CONTRIBUTIONS Andrew Cwiek: Conceptualization; Data curation; Formal analysis; Investigation; Methodol- ogy; Writing – original draft; Writing – review & editing. Sarah Rajtmajer: Conceptualization; Investigation; Methodology; Resources; Writing – original draft; Writing – review & editing. Bradley Wyble: Conceptualization; Resources; Writing – original draft; Writing – review & editing. Vasant Honavar: Resources; Writing – original draft; Writing – review & editing. Emily Grossner: Formal analysis; Investigation; Writing – review & editing. Frank Hillary: Conceptu- alization; Data curation; Formal analysis; Investigation; Methodology; Project administration; Visualization; Writing – original draft; Writing – review & editing. REFERENCES Abadi, M., Barham, P., Chen, J., Chen, Z., Davis, A., Dean, J., ... Baldi, P., Baldi, P. P., Brunak, S., & Bach, F. (2001). Bioinformatics: Zheng, X. (2016). Tensorflow: A system for large-scale machine The machine learning approach. Cambridge, MA: MIT Press. learning. In 12th USENIX symposium on operating systems Baltrušaitis, T., Ahuja, C., & Morency, L. P. (2019). Multimodal design and implementation (OSDI 16) (pp. 265–283). machine learning: A survey and taxonomy. IEEE Transactions Abraham, A., Milham, M. P., Di Martino, A., Craddock, R. C., on Pattern Analysis and Machine Intelligence, 41(2), 423–443. Samaras, D., Thirion, B., & Varoquaux, G. (2017). Deriving https://doi.org/10.1109/TPAMI.2018.2798607 reproducible biomarkers from multi-site resting-state data: An Bassett, D. S., Cullen, K. E., Eickhoff, S. B., Farah, M. J., Goda, Y., Autism-based example. NeuroImage, 147, 736–745. https://doi Haggard, P., ... Ueda, H. R. (2020). Reflections on the past two .org/10.1016/j.neuroimage.2016.10.045, PubMed: 27865923 decades of neuroscience. Nature Reviews Neuroscience, 21(10), Abraham, A., Pedregosa, F., Eickenberg, M., Gervais, P., Mueller, A., 524–534. https://doi.org/10.1038/s41583-020-0363-6, PubMed: Kossaifi, J., ... Varoquaux, G. (2014). Machine learning for neu- 32879507 roimaging with scikit-learn. Frontiers in Neuroinformatics, 8, 14. Bishop, C. M. (2006). Pattern recognition and machine learning.CERN https://doi.org/10.3389/fninf.2014.00014, PubMed: 24600388 Document Server. Springer. https://cds.cern.ch/record/998831 Badea, L., Onu, M., Wu, T., Roceanu, A., & Bajenaru, O. (2017). Bonnelle, V., Ham, T. E., Leech, R., Kinnunen, K. M., Mehta, M. A., Exploring the reproducibility of functional connectivity alterations Greenwood, R. J., & Sharp, D. J. (2012). Salience network integ- in Parkinson’s disease. PLOS ONE, 12(11), e0188196. https://doi rity predicts default mode network function after traumatic brain .org/10.1371/journal.pone.0188196, PubMed: 29182621 injury. Proceedings of the National Academy of Sciences of the Network Neuroscience 44 Predictive modeling and network neuroscience United States of America, 109(12), 4690–4695. https://doi.org/10 patterns of task-defined brain networks allow individual predic- .1073/pnas.1113455109, PubMed: 22393019 tion of cognitive symptom dimension of schizophrenia and are Botvinik-Nezer, R., Holzmeister, F., Camerer, C. F., Dreber, A., linked to molecular architecture. Biological Psychiatry, 89(3), Huber, J., Johannesson, M., Kirchler, M., Iwanir, R., Mumford, 308–319. https://doi.org/10.1016/j.biopsych.2020.09.024, J. A., Adcock, R. A., Avesani, P., Baczkowski, B. M., Bajracharya, PubMed: 33357631 A., Bakst, L., Ball, S., Barilari, M., Bault, N., Beaton, D., Beitner, Chen, L., Chen, Y., Zheng, H., Zhang, B., Wang, F., Fang, J., Li, Y., J., … Schonberg, T. (2020). Variability in the analysis of a single Chen, Q., & Zhang, S. (2020). Changes in the topological orga- neuroimaging dataset by many teams. Nature, 582(7810), 84–88. nization of the default mode network in autism spectrum disor- https://doi.org/10.1038/s41586-020-2314-9,PubMed: der. Brain Imaging and Behavior. https://doi.org/10.1007/s11682 32483374 -020-00312-8, PubMed: 32737824 Bradski, G., & Kaehler, A. (2008). Learning OpenCV: Computer de Filippis, R., Carbone, E. A., Gaetano, R., Bruni, A., Pugliese, V., vision with the OpenCV Library.O’Reilly Media. Segura-Garcia, C., & De Fazio, P. (2019). Machine learning tech- Braun, U., Schaefer, A., Betzel, R. F., Tost, H., Meyer-Lindenberg, niques in a structural and functional MRI diagnostic approach in A., & Bassett, D. S. (2018). From maps to multi-dimensional net- schizophrenia: A systematic review. Neuropsychiatric Disease work mechanisms of mental disorders. Neuron, 97(1), 14–31. and Treatment, 15,1605–1627. https://doi.org/10.2147/NDT https://doi.org/10.1016/j.neuron.2017.11, PubMed: 29301099 .S202418, PubMed: 31354276 Brown, M. R. G, Sidhu, G. S., Greiner, R., Asgarian, N., Bastani, M., Deng, J., Dong, W., Socher, R., Li, L., Li, K., & Fei-Fei, L.. (2009). Silverstone, P. H., Greenshaw, A. J., & Dursun, S. M. (2012). ImageNet: A large-scale hierarchical image database. 2009 ADHD-200 Global Competition: Diagnosing ADHD using IEEE Conference on Computer Vision and Pattern Recognition, personal characteristic data can outperform resting state fMRI 248–255. https://doi.org/10.1109/CVPR.2009.5206848 measurements. Frontiers in Systems Neuroscience, 6. https://doi Dietterich, T. (1995). Overfitting and undercomputing in machine .org/10.3389/fnsys.2012.00069, PubMed: 23060754 learning. ACM Computing Surveys, 27(3), 326–327. https://doi Bullmore, E.T., & Bassett,D.S.(2011).Brain graphs:Graphical .org/10.1145/212094.212114 models of the human brain connectome. Annual Review of Clin- Di Martino, A.,Yan, C.G., Li,Q., Denio, E., Castellanos, F. X., ical Psychology, 7,113–140. https://doi.org/10.1146/annurev Alaerts, K., … Milham, M. P. (2014). The autism brain imaging -clinpsy-040510-143934, PubMed: 21128784 data exchange: Towards a large-scale evaluation of the intrinsic Bullmore, E. T., & Sporns, O. (2009). Complex brain networks: brain architecture in autism. Molecular Psychiatry, 19(6), 659–667. Graph theoretical analysis of structural and functional systems. https://doi.org/10.1038/mp.2013.78,PubMed: 23774715 Nature Reviews Neuroscience, 10(3), 186–198. https://doi.org Esteban, O., Markiewicz, C. J., Blair, R. W., Moodie, C. A., Isik, /10.1038/nrn2575, PubMed: 19190637 A. I.,Erramuzpe,A., … Gorgolewski, K. J. (2019). fMRIPrep: A Bzdok, D., & Ioannidis, J. P. A. (2019). Exploration, inference, and robust preprocessing pipeline for functional MRI. Nature prediction in neuroscience and biomedicine. Trends in Neurosci- Methods, 16(1), 111–116. https://doi.org/10.1038/s41592-018 ences, 42(4), 251–262. https://doi.org/10.1016/j.tins.2019.02 -0235-4, PubMed: 30532080 .001, PubMed: 30808574 Fan, L., Su, J., Qin, J., Hu, D., & Shen, H. (2020). A deep network Caeyenberghs, K., Verhelst, H., Clemente, A., & Wilson, P. H. model on dynamic functional connectivity with applications to (2017). Mapping the functional connectome in traumatic brain gender classification and intelligence prediction. Frontiers in injury: What can graph metrics tell us? NeuroImage, 160,113–123. Neuroscience, 14,881. https://doi.org/10.3389/fnins.2020 https://doi.org/10.1016/j.neuroimage.2016.12.003, PubMed: .00881, PubMed: 33013292 Fawcett, C., & Hoos, H. H. (2016). Analysing differences between Caruana, R., Lawrence, S., & Giles, C. (2000). Overfitting in neu- algorithm configurations through ablation. Journal of Heuristics, ral nets: Backpropagation, conjugate gradient, and early stop- 22(4), 431–458. https://doi.org/10.1007/s10732-014-9275-9 ping. Advances in Neural Information Processing Systems, 13, Forsyth, D., & Ponce, J. (2002). Computer vision: A modern 402–408. approach. Englewood Cliffs, NJ: Prentice-Hall. Cawley, G. C., & Talbot, N. L. C. (2010). On over-fitting in model Gao, J., Li, P., Chen, Z., & Zhang, J. (2020). A survey on deep selection and subsequent selection bias in performance evalua- learning for multimodal data fusion. Neural Computation, 32(5), tion. Journal of Machine Learning Research, 11(70), 2079–2107. 829–864. https://doi.org/10.1162/neco_a_01273, PubMed: Chandrashekar, G., & Sahin, F. (2014). A survey on feature selec- 32186998 tion methods. Computers & Electrical Engineering, 40(1), 16–28. Gelman, A., & Loken, E. (2014). The statistical crisis in science. https://doi.org/10.1016/j.compeleceng.2013.11.024 American Scientist, 102(6), 460–465. https://doi.org/10.1511 /2014.111.460 Chen, F., Wang, Y.-C., Wang, B., & Kuo, C.-C. J. (2020). Graph representation learning: A survey. APSIPA Transactions on Signal Gilbert, N., Bernier, R. A., Calhoun, V. D., Brenner, E., Grossner, E., and Information Processing, 9. https://doi.org/10.1017/ATSIP Rajtmajer, S. M., & Hillary, F. G. (2018). Diminished neural net- .2020.13 work dynamics after moderate and severe traumatic brain injury. PLoS One, 13(6), e0197419. https://doi.org/10.1371/journal Chen, J., Müller, V. I., Dukart, J., Hoffstaedter, F., Baker, J. T., .pone.0197419, PubMed: 29883447 Holmes, A. J., Vatansever, D., Nickl-Jockschat, T., Liu, X., Derntl, B., Kogler, L., Jardri, R., Gruber, O., Aleman, A., Sommer, I. E., Glerean, E., Pan, R. K., Salmi, J., Kujala, R., Lahnakoski, J. M., Roine, Eickhoff, S. B., & Patil, K. R. (2021). Intrinsic connectivity U., Nummenmaa, L., Leppämäki, S., Nieminen-von Wendt, T., Tani, Network Neuroscience 45 Predictive modeling and network neuroscience P., Saramäki, J., Sams, M., & Jääskeläinen, I. P. (2016). Reorganization C., Dale, A. M., Felmlee, J. P., Gunter, J. L., Hill, D. L. G, Killiany, of functionally connected brain subnetworks in high-functioning R., Schuff, N., Fox-Bosetti, S., Lin, C., Studholme, C., … Weiner, autism. Human Brain Mapping, 37(3), 1066–1079. https://doi.org M. W. (2008). The Alzheimer’s disease neuroimaging initiative /10.1002/hbm.23084,PubMed: 26686668 (ADNI): MRI methods. Journal of Magnetic Resonance Imaging, Goodfellow, I., Bengio, Y., & Courville, A. (2016). Deep learning 27(4), 685–691. https://doi.org/10.1002/jmri.21049,PubMed: (illustrated edition). Cambridge, MA: The MIT Press. 18302232 Guo, W., Wang, J., & Wang, S. (2019). Deep multimodal represen- Kosslyn, S. M., Thompson, W. L., Kim, I. J., & Alpert, N. M. (1995). tation learning: A survey. IEEE Access, 7, 63373–63394. https:// Topographical representations of mental images in primary visual doi.org/10.1109/ACCESS.2019.2916887 cortex. Nature, 378(6556), 496–498. https://doi.org/10.1038 Guyon, I., & Elisseeff, A. (2003). An introduction to variable and /378496a0, PubMed: 7477406 feature selection. The Journal of Machine Learning Research, 3, Kraguljac, N. V., McDonald, W. M., Widge, A. S., Rodriguez, C. I., 1157–1182. Tohen, M., & Nemeroff, C. B. (2021). Neuroimaging biomarkers Hallquist, M. N., & Hillary, F. G. (2018). Graph theory approaches in schizophrenia. The American Journal of Psychiatry. https://doi to functional network organization in brain disorders: A critique .org/10.1176/appi.ajp.2020.20030340, PubMed: 33397140 for a brave new small-world. Network Neuroscience, 3(1), 1–26. Langella, S., Sadiq, M. U., Mucha, P. J., Giovanello, K. S., Dayan, https://doi.org/10.1162/netn_a_00054, PubMed: 30793071 E., & Alzheimer’s Disease Neuroimaging Initiative. (2021). Lower Hamilton, W. L., Ying, R., & Leskovec, J. (2017). Representation functional hippocampal redundancy in mild cognitive impair- learning on graphs: Methods and applications. arXiv. https:// ment. Translational Psychiatry, 11(1), 61. https://doi.org/10.1038 arxiv.org/abs/1709.05584v3 /s41398-020-01166-w, PubMed: 33462184 Hastie, T., Tibshirani, R., & Friedman, J. (2009). The elements of LaPlaca, M. C., Huie, J. R., Alam, H. B., Bachstetter, A. D., Bayir, statistical learning: Data mining, inference, and prediction H., Bellgowan, P. S. F, Cummings, D., Dixon, C. E., Ferguson, (2nd ed.). Springer Science & Business Media. https://doi.org A. R., Ferland-Beckham, C., Floyd, C., Friess, S., Galanopoulou, /10.1007/978-0-387-84858-7 A., Hall, E. D., Harris, N. G., Hawkins, B. E., Hicks, R., Hulbert, Hawkins, D. M. (2004). The problem of overfitting. Journal of L. E., Johnson, V. E., … Zai, L. (2020). Preclinical common data Chemical Information and Computer Sciences, 44(1), 1–12. elements for traumatic brain injury research: Progress and use https://doi.org/10.1021/ci0342472, PubMed: 14741005 cases. Journal of Neurotrauma. https://doi.org/10.1089/neu Hedderich, D. M., & Eickhoff, S. B. (2021). Machine learning for .2020.7328, PubMed: 33297844 psychiatry: Getting doctors at the black box? Molecular Psychia- Larrañaga, P., Calvo, B., Santana, R., Bielza, C., Galdiano, J., Inza, try, 26(1), 23–25. https://doi.org/10.1038/s41380-020-00931-z, I., Lozano, J. A., Armañanzas, R., Santafé, G., Pérez, A., & Robles, PubMed: 33173196 V. (2006). Machine learning in bioinformatics. Briefings in Bioin- Hegarty, J. P., Ferguson, B. J., Zamzow, R. M., Rohowetz, L. J., Johnson, formatics, 7(1), 86–112. https://doi.org/10.1093/bib/bbk007, J. D., Christ, S. E., & Beversdorf, D. Q. (2017). Beta-adrenergic PubMed: 16761367 antagonism modulates functional connectivity in the default Le, T., & Honavar, V. (2020). Dynamical Gaussian process latent mode network of individuals with and without autism spectrum variable model for representation learning from longitudinal disorder. Brain Imaging and Behavior, 11(5), 1278–1289. https:// data. Proceedings of the 2020 ACM-IMS on Foundations of Data doi.org/10.1007/s11682-016-9604-8,PubMed: 27714553 Science Conference, 183–188. https://doi.org/10.1145/3412815 Heinrichs, B., & Eickhoff, S. B. (2020). Your evidence? Machine learning algorithms for medical diagnosis and prediction. Human LeCun, Y., Bengio, Y., & Hinton, G. (2015). Deep learning. Nature, Brain Mapping, 41(6), 1435–1444. https://doi.org/10.1002/hbm 521(7553), 436–444. https://doi.org/10.1038/nature14539, .24886, PubMed: 31804003 PubMed: 26017442 Hiremath, C. S., Sagar, K. J. V, Yamini, B. K., Girimaji, A. S., Kumar, Lefort-Besnard, J., Bassett, D. S., Smallwood, J., Margulies, D. S., R., Sravanti, S. L., Padmanabha, H., Vykunta Raju, K. N., Kishore, Derntl, B., Gruber, O., Aleman, A., Jardri, R., Varoquaux, G., M. T., Jacob, P., Saini, J., Bharath, R. D., Seshadri, S. P., & Kumar, Thirion, B., Eickhoff, S. B., & Bzdok, D. (2018). Different shades M. (2021). Emerging behavioral and neuroimaging biomarkers for of default mode disturbance in schizophrenia: Subnodal covari- early and accurate characterization of autism spectrum disorders: ance estimation in structure and function. Human Brain Map- A systematic review. Translational Psychiatry, 11(1), 42. https://doi ping, 39(2), 644–661. https://doi.org/10.1002/hbm.23870, .org/10.1038/s41398-020-01178-6, PubMed: 33441539 PubMed: 29105239 Hosseini, M., Powell, M., Collins, J., Callahan-Flintoft, C., Jones, W., Liang, J., Wu, Y., Xu, D., & Honavar, V. (2020a). Longitudinal deep Bowman, H., & Wyble, B. (2020). I tried a bunch of things: The kernel Gaussian process regression. ArXiv. https://arxiv.org/abs dangers of unexpected overfitting in classification of brain data. /2005.11770 Neuroscience and Biobehavioral Reviews, 119, 456–467. https:// Liang, J., Xu, D., Sun, Y., & Honavar, V. (2020b). LMLFM: longitu- doi.org/10.1016/j.neubiorev.2020.09.036, PubMed: 33035522 dinal multi-level factorization machine. Proceedings of the AAAI Hsieh, T.-Y., Wang, S., Sun, Y., & Honavar, V. (2020). Explainable Conference on Artificial Intelligence, 34(04), 4811–4818. https:// multivariate time series classification: A deep neural network doi.org/10.1609/aaai.v34i04.5916 which learns to attend to important variables as well as informa- Lu, H., Liu, S., Wei, H., Chen, C., & Geng, X. (2021). Deep tive time intervals. ArXiv. https://arxiv.org/abs/2011.11631 multi-kernel auto-encoder network for clustering brain functional Jack, C. R., Bernstein, M. A., Fox, N. C., Thompson, P., Alexander, connectivity data. Neural Networks, 135,148–157. https://doi.org G., Harvey, D., Borowski, B., Britson, P. J., Whitwell, J. L., Ward, /10.1016/j.neunet.2020.12.005,PubMed: 33388506 Network Neuroscience 46 Predictive modeling and network neuroscience Manning, C., Surdeanu, M., Bauer, J., Finkel, J., Bethard, S., & Standardization, 16(2), 119–127. https://doi.org/10.1016/0092 McClosky, D. (2014). The Stanford CoreNLP Natural Language -1157(88)90039-x, PubMed: 3131345 Processing Toolkit. Proceedings of 52nd Annual Meeting of the Radua, J., Vieta, E., Shinohara, R., Kochunov, P., Quidé, Y., Green, Association for Computational Linguistics: System Demonstra- M. J., Weickert, C. S., Weickert, T., Bruggemann, J., Kircher, T., tions,55–60. https://doi.org/10.3115/v1/P14-5010 Nenadić, I., Cairns, M. J., Seal, M., Schall, U., Henskens, F., Markiewicz, C. J.,Gorgolewski,K.J., Feingold, F.,Blair,R., Fullerton, J. M., Mowry, B., Pantelis, C., Lenroot, R., … van Erp, Halchenko, Y. O., Miller, E., … Poldrack, R. A. (2021). Open- T. (2020). Increased power by harmonizing structural MRI site Neuro: An open resource for sharing of neuroimaging data. differences with the ComBat batch adjustment method in bioRxiv. https://doi.org/10.1101/2021.06.28.450168 ENIGMA. NeuroImage, 218, 116956. https://doi.org/10.1016/j Mei, J., Desrosiers, C., & Frasnelli, J. (2020). Machine learning for .neuroimage.2020.116956, PubMed: 32470572 the diagnosis of Parkinson’s disease: A systematic review. ArXiv. Roelofs, R., Shankar, V., Recht, B., Fridovich-Keil, S., Hardt, M., https://arxiv.org/abs/2010.06101 Miller, J., & Schmidt, L. (2019). A meta-analysis of overfit- Meyes, R., Lu, M., de Puiseau, C. W., & Meisen, T. (2019). Ablation ting in machine learning. In Proceedings of the 33rd Inter- studies in artificial neural networks. ArXiv. https://arxiv.org/abs national Conference on Neural Information Processing Systems /1901.08644 (pp. 9179–9189). Mikolov, T., Chen, K., Corrado, G., & Dean, J. (2013). Efficient esti- Roy, A., Bernier, R. A., Wang, J., Benson, M., French, J. J., Good, mation of word representations in vector space. ArXiv. https:// D. C., & Hillary, F. G. (2017). The evolution of cost-efficiency arxiv.org/abs/1301.3781 in neural networks during recovery from traumatic brain injury. Min, S., Lee, B., & Yoon, S. (2017). Deep learning in bioinformatics. PLoS One, 12(4), e0170541. https://doi.org/10.1371/journal Briefings in Bioinformatics, 18(5), 851–869. https://doi.org/10 .pone.0170541, PubMed: 28422992 Rubbert, C., Mathys, C., Jockwitz, C., Hartmann, C. J., Eickhoff, S. B., .1093/bib/bbw068, PubMed: 27473064 Hoffstaedter, F., Caspers, S., Eickhoff, C. R., Sigl, B., Teichert, Mitchell, T. M. (1997). Machine learning (1st ed.). McGraw-Hill. N. A., Südmeyer, M., Turowski, B., Schnitzler, A., & Caspers, J. Ng, A. (1997). Preventing “overfitting” of cross-validation data. (2019). Machine-learning identifies Parkinson’s disease patients International Conference on Machine Learning, 97, 245–253. Nichols, T. E., Das, S., Eickhoff, S. B., Evans, A. C., Glatard, T., based on resting-state between-network functional connectivity. Hanke, M., Kriegeskorte, N., Milham, M. P., Poldrack, R. A., The British Journal of Radiology, 92(1101), 20180886. https:// Poline, J.-B., Proal, E., Thirion, B., Van Essen, D. C., White, T., & doi.org/10.1259/bjr.20180886, PubMed: 30994036 Yeo, B. T. T (2017). Best practices in data analysis and sharing in Salvatore, C., Cerasa, A., Battista, P., Gilardi, M. C., Quattrone, A., neuroimaging using MRI. Nature Neuroscience, 20(3), 299–303. & Castiglioni, I. (2015). Magnetic resonance imaging biomarkers https://doi.org/10.1038/nn.4500, PubMed: 28230846 for the early diagnosis of Alzheimer’s disease: A machine learn- Page, M. J., McKenzie, J. E., Bossuyt, P. M., Boutron, I., Hoffmann, T. C., ing approach. Frontiers in Neuroscience, 9. https://doi.org/10 Mulrow, C. D., Shamseer, L., Tetzlaff, J. M., Akl, E. A., Brennan, S. E., .3389/fnins.2015.00307, PubMed: 26388719 Chou, R., Glanville, J., Grimshaw, J. M., Hróbjartsson, A., Lalu, Sarwar, T., Seguin, C., Ramamohanarao, K., & Zalesky, A. (2020). M. M., Li, T., Loder, E. W., Mayo-Wilson, E., McDonald, S., … Towards deep learning for connectome mapping: A block decom- Moher, D. (2021). The PRISMA 2020 statement: An updated position framework. NeuroImage, 212, 116654. https://doi.org/10 guideline for reporting systematic reviews. The BMJ, 372, n71. .1016/j.neuroimage.2020.116654, PubMed: 32068163 https://doi.org/10.1136/bmj.n71, PubMed: 33782057 Schaffer, C. (1993). Overfitting avoidance as bias. Machine Learning, Parkes, L., Satterthwaite, T. D., & Bassett, D. S. (2020). Towards pre- 10(2), 153–178. https://doi.org/10.1007/BF00993504 cise resting-state fMRI biomarkers in psychiatry: Synthesizing Sheikholeslami, S. (2019). Ablation programming for machine developments in transdiagnostic research, dimensional models learning. DiVA. diva2:1349978. of psychopathology, and normative neurodevelopment. Current Silva, R. F., Castro, E., Gupta, C. N., Cetin, M., Arbabshirani, M., Opinion in Neurobiology, 65, 120–128. https://doi.org/10.1016 Potluru, V. K., Plis, S. M., & Calhoun, V. D. (2014). The tenth /j.conb.2020.10.016, PubMed: 33242721 annual MLSP competition: Schizophrenia classification chal- Pellegrini, E., Ballerini, L., Hernandez, M. del C. V., Chappell, lenge. 2014 IEEE International Workshop on Machine Learning F. M., González-Castro, V., Anblagan, D., Danso, S., Muñoz- for Signal Processing (MLSP),1–6. https://doi.org/10.1109/MLSP Maniega, S., Job, D., Pernet, C., Mair, G., MacGillivray, T. J., .2014.6958889 Trucco, E., & Wardlaw, J. M. (2018). Machine learning of neuro- Srivastava, N., Hinton, G., Krizhevsky, A., Sutskever, I., & Salakhutdinov, imaging for assisted diagnosis of cognitive impairment and R. (2014). Dropout: A simple way to prevent neural networks dementia: A systematic review. Alzheimer’s & Dementia: Diag- from overfitting. The Journal of Machine Learning Research, 15(1), nosis, Assessment & Disease Monitoring, 10, 519–535. https:// 1929–1958. doi.org/10.1016/j.dadm.2018.07.004, PubMed: 30364671 Steardo, L., Carbone, E. A., de Filippis, R., Pisanu, C., Segura-Garcia, Poldrack, R. A., Huckins, G., & Varoquaux, G. (2020). Establish- C., Squassina, A., De Fazio, P., & Steardo, L. (2020). Application ment of best practices for evidence for prediction: A review. of support vector machine on fMRI data as biomarkers in schizo- JAMA Psychiatry, 77(5), 534–540. https://doi.org/10.1001 phrenia diagnosis: A systematic review. Frontiers in Psychiatry, /jamapsychiatry.2019.3671, PubMed: 31774490 11, 588. https://doi.org/10.3389/fpsyt.2020.00588, PubMed: Progar, J. J., & May, J. C. (1988). The determination of low levels of 32670113 aluminum in antihemophilic factor (human) preparations by Thompson, P. M., Jahanshad, N., Ching, C. R. K, Salminen, L. E., flame atomic absorption spectrometry. Journal of Biological Thomopoulos, S. I., Bright, J., Baune, B. T., Bertolín, S., Bralten, Network Neuroscience 47 Predictive modeling and network neuroscience J., Bruin, W. B., Bülow, R., Chen, J., Chye, Y., Dannlowski, U., de using graph theory. PLoS One, 5(10), e13701. https://doi.org/10 Kovel, C. G. F, Donohoe, G., Eyler, L. T., Faraone, S. V., Favre, P., .1371/journal.pone.0013701, PubMed: 21060892 … ENIGMA Consortium. (2020). ENIGMA and global neurosci- Varoquaux, G., Raamana, P. R., Engemann, D. A., Hoyos-Idrobo, ence: A decade of large-scale studies of the brain in health and A., Schwartz, Y., & Thirion, B. (2017). Assessing and tuning brain disease across more than 40 countries. Translational Psychiatry, decoders: Cross-validation, caveats, and guidelines. Neuro- 10(1), 100. https://doi.org/10.1038/s41398-020-0705-1, Image, 145, 166–179. https://doi.org/10.1016/j.neuroimage PubMed: 32198361 .2016.10.038, PubMed: 27989847 Thompson, P. M., Stein, J. L., Medland, S. E., Hibar, D. P., Vasquez, Vu, M. A. T., Adali, T., Ba, D., Buzsáki, G., Carlson, D., Heller, K., ... A. A., Renteria, M. E., Toro, R., Jahanshad, N., Schumann, G., Dzirasa, K. (2018). A shared vision for machine learning in neu- Franke, B., Wright, M. J., Martin, N. G., Agartz, I., Alda, M., roscience. Journal of Neuroscience, 38(7), 1601–1607. https://doi Alhusaini, S., Almasy, L., Almeida, J., Alpert, K., Andreasen, .org/10.1523/JNEUROSCI.0508-17.2018, PubMed: 29374138 N. C., … Alzheimer’s Disease Neuroimaging Initiative, EPIGEN Vul, E., Harris, C., Winkielman, P., & Pashler, H. (2009). Puzzlingly Consortium, IMAGEN Consortium, Saguenay Youth Study (SYS) high correlations in fMRI studies of emotion, personality, and Group. (2014). The ENIGMA Consortium: Large-scale collabora- social cognition. Perspectives on Psychological Science, 4(3), tive analyses of neuroimaging and genetic data. Brain Imaging 274–290. https://doi.org/10.1111/j.1745-6924.2009.01125.x, and Behavior, 8(2), 153–182. https://doi.org/10.1007/s11682 PubMed: 26158964 -013-9269-5, PubMed: 24399358 Yarkoni, T., & Westfall, J. (2017). Choosing prediction over expla- Turian, J., Ratinov, L.-A., & Bengio, Y. (2010). Word representations: nation in psychology: Lessons from machine learning. Perspec- A simple and general method for semi-supervised learning. In tives on Psychological Science: A Journal of the Association for Proceedings of the 48th Annual Meeting of the Association for Psychological Science, 12(6), 1100–1122. https://doi.org/10 Computational Linguistics (pp. 384–394). .1177/1745691617693393, PubMed: 28841086 van den Heuvel, M. P., de Lange, S. C., Zalesky, A., Seguin, C., Yeo, Zhang-James, Y., Hoogman, M., Franke, B., & Faraone, S. V. (2020). B. T. T., & Schmidt, R. (2017). Proportional thresholding in Machine learning and MRI-based diagnostic models for ADHD: resting-state fMRI functional connectivity networks and conse- Arewethere yet? MedRxiv, 2020.10.20.20216390. https://doi quences for patient-control connectome studies: Issues and rec- .org/10.1101/2020.10.20.20216390 ommendations. NeuroImage, 152,437–449. https://doi.org/10 Zuo, X.-N., Biswal, B. B., & Poldrack, R. A. (2019). Editorial: Reli- .1016/j.neuroimage.2017.02.005, PubMed: 28167349 ability and reproducibility in functional connectomics. Frontiers van Wijk, B. C. M., Stam, C. J., & Daffertshofer, A. (2010). Compar- in Neuroscience, 13,117. https://doi.org/10.3389/fnins.2019 ing brain networks of different size and connectivity density .00117, PubMed: 30842722 Network Neuroscience 48

Journal

Network Neuroscience – MIT Press

Published: Feb 1, 2022

There are no references for this article.