Calculation of exact Shapley values for explaining support vector machine models using the radial basis function...

Area of Science:

• Computational chemistry
• Artificial intelligence
• Drug discovery

Background:

• Machine learning (ML) models are crucial in pharmaceutical research but often act as black boxes, hindering interpretation.
• Explainable AI (XAI) methods, particularly Shapley values, aim to rationalize ML predictions for experimental design.
• Existing approximations like SHAP show limited correlation with exact Shapley values for Support Vector Machine (SVM) models, especially with Tanimoto kernels.

Purpose of the Study:

• To develop a computationally efficient method for calculating exact Shapley values for SVM models.
• To address the limitations of SHAP in explaining SVM predictions accurately.
• To enable more reliable interpretation of ML models in pharmaceutical research.

Main Methods:

• Developed the Shapley value-expressed radial basis function (SVERAD) approach.
• Applied SVERAD to calculate exact Shapley values for SVM models using radial basis function kernels.
• Evaluated SVERAD's efficiency and accuracy compared to existing methods.

Main Results:

• SVERAD enables efficient computation of exact Shapley values for SVM models.
• The new method provides meaningful explanations for SVM predictions.
• Demonstrated improved correlation between SVERAD and exact Shapley values compared to SHAP for SVMs.

Conclusions:

• SVERAD offers a computationally efficient and accurate solution for explaining SVM models in pharmaceutical research.
• This advancement facilitates better understanding and trust in ML-driven predictions.
• The method supports the rational design of experiments in drug discovery.