Druggability of targets is crucial in drug target selection. This paper presents DrugnomeAI, a stochastic semi-supervised machine learning framework that estimates druggability likelihood for every protein-coding gene in the human exome. Integrating gene-level properties from 15 sources (324 features), DrugnomeAI generates exome-wide predictions (median AUC: 0.97), highlighting protein-protein interaction networks as top predictors. The tool offers generic and specialized models stratified by disease type or drug modality. Top-ranking genes showed significant enrichment for genes in clinical development programs (p<1×10^-308) and genome-wide significance in UK Biobank exome-wide association studies (p=1.7×10^-5 for binary and p=1.6×10^-7 for quantitative traits). A web application (http://drugnomeai.public.cgr.astrazeneca.com) visualizes predictions and key features.

Arwa Raies, Ewa Tulodziecka, James Stainer, Lawrence Middleton, Ryan S. Dhindsa, Pamela Hill, Ola Engkvist, Andrew R. Harper, Slavé Petrovski, Dimitrios Vitsios