Clinically distinct genetic diseases converge on shared, treatable nodes, as AI-powered discovery tool shows. This tool identifies drug-targetable nodes and accelerates the development of clinical targets for the treatment of genetic diseases. The study integrates multiple lines of evidence from human genetics in a probabilistic framework to systematically prioritize drug targets. Using network diffusion on PPI networks and biological functions, he explains how drugs treat diseases by targeting impaired functions without directly modulating disease-associated proteins. TxGNN uses graph neural networks for zero-shot drug repurposing, where drugs and diseases are embedded in a shared latent space to identify similar diseases by molecular, phenotypic, or pathway similarities. The proteogenomic map includes 1859 gene-protein-phenotype associations with 412 proteins and 506 curatorial features, highlighting strong biological convergence across diseases. Of the 1538 protein targets with cis-pQTL, half share a genetic signal with gene expression in at least one of the 49 tissues. Computational methods are urgently needed to accelerate target discovery for rare diseases.