Calculation of Exact Shapley Values for Support Vector Machines with Tanimoto Kernel Enables Model Interpretation

Code and analysis workflow for calculating Shapley values (SVs) for support vector machines (SVM) with Tanimoto kernel.

The Shapley Value-Expressed Tanimoto similarity (SVETA) approach enables the efficient calculation of SVs for SVM predictions. Code provided in the folder 'sveta' contains functions for the computation of SVETA values and SVM algorithm to explain individual predictions.

Jupyter notebooks containing workflows for assessing SVETA values can be found in the folder 'analysis'.

An exemplary data set consisting of compounds active against the adenosine receptor A3 (UniProt ID: P0DMS8) and random compounds is provided in the folder 'data'.