Abstract
Determining the correct direction of effect (DOE), whether to increase or decrease the activity of a drug target, is essential for therapeutic success. We introduce a framework to predict DOE at gene and gene-disease levels using gene and protein embeddings and genetic associations across the allele frequency spectrum, respectively. Specifically, we predict: (1) DOE-specific druggability for 19,450 protein-coding genes with a macro-averaged area under the receiver operating characteristic curve (AUROC) of 0.95; (2) isolated DOE among 2553 druggable genes with a macro-averaged AUROC of 0.85; and (3) gene-disease-specific DOE for 47,822 gene-disease pairs with a macro-averaged AUROC of 0.59, with performance improving with genetic evidence availability. Our predictions outperform existing approaches, are associated with clinical trial success, and identify novel therapeutic opportunities. We uncover genetic and functional differences between activator and inhibitor targets, allowing DOE inference independent of disease context. This framework represents a valuable tool for target selection and drug development.