Abstract

One of the first challenges in understanding neural information processing is the identification of the functional roles of neural network elements. Aiming at this goal, lesion studies have been classically used in neuroscience, most of which have employed single lesions which are limited in their ability to reveal the significance of interacting elements. The recently developed Functional Contribution Analysis (FCA) method has addressed the functional localization challenge by analyzing data composed of multiple lesioning experiments and corresponding functional performance levels, using an operative minimization approach. This paper presents the Multi-lesion Shapley value Ana/ysis (MSA), an axiomatic, scalable and rigorous method for deducing causal function localization from multiple lesioning data, overcoming several shortcomings of the FCA. The MSA, based on fundamental concepts from game theory, accurately quantifies the contributions of network elements and their interactions. While the original game theoretical definition and calculation of the Shapley value requires a data set of a potentially vast number of all multiple lesion experiments, we developed several MSA prediction and estimation variants which use only a relatively small set of experiments. The successful working of the MSA is demonstrated in a theoretical test case, in artificially evolved neurocontrollers and for the analysis of an example of biological, reversible deactivation data. MSA has a wide range of potential applications in neuroscience for the analysis of reversible deactivation experiments and transcranial magnetic stimulation "virtual lesions", and in biology in general, for the analysis of gene networks via "multi-knockout" experiments.

Citations