Learning, explaining and actioning general principles of cancer cell drug sensitivity

Presenter: Francesco Carli

Location: Officine Garibaldi - il Cantiere delle Idee, Via Vincenzo Gioberti, 39, 56124 Pisa PI, Italia

Abstract. Unlocking the potential of large pharmacogenomics datasets for targeted cancer therapy requires both predictive accuracy and interpretability. This talk delves into a novel framework integrating explainable XAI techniques, such as SHAP and permutation importance, with traditional bioinformatics methods and large language models (LLMs). The approach enhances the understanding of drug mechanisms of action (MOAs) by linking enriched pathways to drug-specific gene importance. Demonstrations show how these insights can predict cell lines drug sensitivity, highlight critical biological processes, and identify potential new therapeutic targets, offering a robust pipeline for translating in vitro findings to clinical settings.