Explainable AI for Chemotherapy Response: A SHAP-Integrated Random Forest Approach

In our work, we propose an explainable AI framework to predict the response to chemotherapy drugs using gene-expression profiles. Our system employs Random Forest regression to simulate the response of 50 anticancer drugs across approximately 2,000 genes alongside SHAP based explanations of the gene-level effects. Using Dactolisib (928 samples) as a representative example, this study's model illustrates an MSE of 1.0406, with an MAE of 0.8049 and R2 of 0.2108. These results outperform Ridge Regression and XGBoost under 5-fold crossvalidation. The findings of the SHAP analysis are biologically meaningful, especially genes relevant to PI3K/mTOR signaling. We combine the model along with a Gradio interface for prediction and explainability in real-time. The framework shows that the model for drug-response is accurate, transparent, and scalable for precision oncology.