drGAT

A challenge in drug response prediction is result interpretation compared to established knowledge. drGAT is a graph deep learning model that predicts sensitivity and aids in biomarker identification using attention coefficients. drGAT leverages a heterogeneous graph of relationships drawn from drugs, proteins, and cell line responses. drGAT demonstrates competitive performance among benchmarks, achieving 76\% F1-score for DNA-damaging compounds from the NCI60 pharmacogenomics dataset. In the validation of independent unseen responses, we achieve around 80\% F1-score. Regarding interpretability, we review drug-gene co-occurrences by text-mining PubMed abstracts for high-coefficient genes mentioning particular drugs. Across 162 drugs with known drug-target interactions (DTIs), our model utilized 95.52\% of known DTIs and suggested additional predictive associations, many supported by the literature. We use attention coefficients to identify affected biological processes by each drug via enrichment analyses.