Interpreting neural decoding models using grouped model reliance

• Published in: PLoS computational biology Vol. 16; no. 1; p. e1007148
• Main Authors: Valentin, Simon, Harkotte, Maximilian, Popov, Tzvetan
• Format: Journal Article
• Language: English
• Published: United States Public Library of Science 01.01.2020; Public Library of Science (PLoS)
• Subjects: Adult; Algorithms; Biology and Life Sciences; Brain - physiology; Brain research; Cognitive ability; Cognitive Neuroscience - methods; Computational Biology - methods; Computer and Information Sciences; Decision Trees; Digits; EEG; Electroencephalography; Electroencephalography - classification; Female; Frequencies; Humans; Learning algorithms; Machine Learning; Male; Medicine and Health Sciences; Memory; Memory tasks; Memory, Short-Term - physiology; Mental task performance; Methods; Models, Neurological; Nervous system; Neural coding; Neurosciences; Permutations; Physical Sciences; Psychology; Research and Analysis Methods; Researchers; Retention; Short term memory; Support Vector Machine; Support vector machines; Task Performance and Analysis; Variables; Young Adult; Germany
• ISSN: 1553-7358; 1553-734X; 1553-7358
• DOI: 10.1371/journal.pcbi.1007148

Machine learning algorithms are becoming increasingly popular for decoding psychological constructs based on neural data. However, as a step towards bridging the gap between theory-driven cognitive neuroscience and data-driven decoding approaches, there is a need for methods that allow to interpret trained decoding models. The present study demonstrates grouped model reliance as a model-agnostic permutation-based approach to this problem. Grouped model reliance indicates the extent to which a trained model relies on conceptually related groups of variables, such as frequency bands or regions of interest in electroencephalographic (EEG) data. As a case study to demonstrate the method, random forest and support vector machine models were trained on within-participant single-trial EEG data from a Sternberg working memory task. Participants were asked to memorize a sequence of digits (0-9), varying randomly in length between one, four and seven digits, where EEG recordings for working memory load estimation were taken from a 3-second retention interval. The present results confirm previous findings insofar as both random forest and support vector machine models relied on alpha-band activity in most subjects. However, as revealed by further analyses, patterns in frequency and topography varied considerably between individuals, pointing to more pronounced inter-individual differences than previously reported.