Semi-empirical and machine learning-based prediction of site of metabolisms mediated by aldehyde oxidase

Prediction of drug clearance by aldehyde oxidase (AO) catalyzed metabolism has been of great interest in drug discovery. We report a fast, accurate, and fully-automated tool that combines semi-empirical quantum mechanical (SQM) calculations and machine-learning (ML) technique to predict AO-mediated site of metabolism (SOM). We infer that calculated probability scores together with those based on approximate C(sp2)-H bond dissociation energies accomplish accurate and immediate prediction of the number and rank ordering of SOMs. Tests on a series of 150 AO substrates demonstrate that our computational strategy can be applicable to high-throughput screening of metabolically stable compounds.