Accelerating Primary Screening of USP8 Inhibitors from Drug Repurposing Databases with Tree‐Based Machine Learning

The ubiquitin‐proteasome system is critical for maintaining protein homeostasis, with deubiquitinases such as ubiquitin‐specific protease 8 (USP8) regulating protein stability through removal of ubiquitin chains. USP8 plays essential roles in signaling, receptor trafficking, and cell cycle control. Its dysregulation has been implicated in the progression of several cancers, including intrahepatic cholangiocarcinoma, breast and hepatocellular carcinomas. Despite promising preclinical results with several small‐molecule USP8 inhibitors, none have advanced to clinical approval, highlighting the challenges of traditional drug development. Drug repurposing offers a cost‐ and time‐efficient alternative by leveraging existing pharmacological and safety data. To accelerate the discovery of USP8 inhibitors, supervised machine learning (ML) techniques are applied to high‐throughput screening data. Six ML architectures are trained and evaluated using several molecular fingerprints, identifying an optimized XGBoost model that achieved a 16.3‐fold improvement in positive hit‐rate over baseline. This model was used to virtually screen compounds from DrugBank and the Broad Repurposing Hub, yielding 100 high‐priority USP8 inhibitor candidates, nearly half of which are already in clinical trials or approved for other indications. Nine distinct molecular scaffolds that has not been studied as USP8 binders were also identified. This integrative approach provides a scalable, efficient pipeline for repurposing‐driven drug discovery targeting USP8.