ProphDR: An Interpretable Deep Learning Model for Predicting Cancer Drug Response via Multi-Omics and Cross-Attention Mechanisms.
Summary
ProphDR is a novel interpretable deep learning model designed to predict cancer drug responses by integrating multi-omics data and drug structural information. The model employs hierarchical attention mechanisms, including a Criss-Cross Gene-level Multiomics Integration (CGMI) module and a cross-attention (CA) module to model drug-gene interactions. Evaluated on GDSC and CCLE datasets, ProphDR achieves state-of-the-art performance in predicting ln(IC50) values and classifying drug sensitivity. It also demonstrates strong generalizability, even for unseen drugs or cell lines, and generates biologically interpretable attention maps.
Analysis
🔴 CLINIQUE: This model represents a significant step forward for precision oncology by offering an interpretable predictive tool. By identifying key resistance-related genes, such as ERBB2, ProphDR could guide therapeutic target prioritization and drug repurposing, particularly in cancers like NSCLC and BRCA. While a computational model, its ability to link genomic features to phenotypic outcomes could justify clinical trials exploring therapeutic strategies based on its predictions, with a potential clinical impact in 5+ years. It acts as a predictive biomarker for treatment response. 🔵 BIOINFO: ProphDR is an innovative deep learning model that integrates multi-omics data and drug structural information through hierarchical and cross-attention mechanisms, representing a sophisticated approach for heterogeneous data integration. The model stands out due to its interpretability, generating attention maps that highlight relevant genes and pharmacophores, which is crucial for clinical trust and adoption. Its state-of-the-art performance (PCC = 0.938, RMSE = 0.978, AUC = 0.981) on the extensive GDSC and CCLE datasets, along with its generalizability in cold-start scenarios, attests to its robustness. Clinically, ProphDR could be deployed to refine treatment selection, prioritize drug candidates for preclinical testing, or assist molecular tumor boards in identifying potential resistance mechanisms, thereby changing decision-making workflows based on genomic and drug data.