Multi-omics Informed Drug Response Prediction for Precision Medicine

Motivation: Treating cancer in patients is hard within precision oncology due to the heterogeneity in cancer cells. Developing predictive methods that estimate drug-responses in cancer cell lines can improve our understanding of treatment effectiveness. The goal of this study is to explore the fundamental and data-driven limitations of the existing advanced learning methods for drug response prediction for precision oncology. Approach & Findings: We examine current state-of the-art methods that utilize multi-omics patient data and/or signaling pathways to predict drug response across various cell lines. Our analysis identifies that models that learn molecular and topological manifolds demonstrate richer representation of drug molecules, and better efficacy of drug-response prediction. We also identify the core design principles that are necessary for developing an effective computational framework. Future work will explore various options in the architecture design principles and attempt to improve drug-response prediction, especially when multi-omics data is not present for a patients.