OTRec: deep learning recommender for prospective druggable disease–target associations

OTRec is a deep learning recommender system that prospectively predicts druggable disease– target associations. Unlike methods that rely on manually aggregated evidence scores, OTRec employs a two-tower architecture to learn latent representations directly from ∼640,000 pairs of disease and target (gene) text descriptions, and biological annotations (e.g., tractability, gene ontology, pathways). Validated temporally to predict 2025 clinical entry from 2022, OTRec significantly outperforms the retrospective Open Targets (OT) association score (ROC-AUC 0.865 vs 0.56) and improves upon state-of-the-art baselines in target-disjoint settings (ROC-AUC 0.949 vs 0.91). We provide ranked druggable genome repurposing candidates for over 17,000 diseases, including 2,322 orphan diseases. Source code and ranked predictions available at https://github.com/LinialLab/OTRec An interactive model is available at https://huggingface.co/spaces/GrimSqueaker/OTRec [email protected]