Topology-Aware Deep Learning on Higher-Order Structures for Drug Response Prediction

Accurate prediction of anticancer drug response remains a central challenge in precision oncology. Existing approaches often rely on pairwise modeling, overlooking higher-order dependencies among drugs and cell lines. We present TopDr, a topology-aware deep learning framework that encodes both drugs and cell lines as multiscale simplicial complexes, capturing interactions at the 0-, 1-, and 2-simplex levels. By jointly integrating local higher-order neighborhoods and global topological structures, TopDr generates enriched representations for sensitivity prediction. Across six benchmark datasets, TopDr consistently matches or surpasses state-of-the-art baselines in both regression and classification tasks. Beyond predictive accuracy, TopDr offers mechanism-level interpretability: attention over 1- and 2-simplices highlights drug pairs and triplets with significant pathway enrichment, while cell line groupings reveal biologically coherent expression modules. These results demonstrate that modeling multiscale higher-order topology yields predictions that are not only accurate and robust but also biologically interpretable, paving the way for more reliable drug response modeling.