Combining machine-learning and molecular-modeling methods for drug-target affinity predictions

Machine learning (ML) techniques offer a novel and exciting approach in the drug discovery field. One might even argue that their current expansion may push traditional MM modeling techniques to a secondary role in modeling methods. In this review article, we advocate that a combination of both techniques could be the most efficient implementation in the coming years. Focusing on drug-target affinity predictions, we first review pure ML approaches. Then, we introduced recent developments in mixing ML and MM methods in a single combined manner. Finally, we show the detailed implementation of a real industrial prospective study where nanomolar hits, on a kinase target, were obtained by combination of state of the art Monte Carlo MM simulations (PELE) with a ML ranking function.This article is categorized under:Structure and Mechanism > Computational Biochemistry and BiophysicsData Science > Artificial Intelligence/Machine LearningMolecular and Statistical Mechanics > Molecular Mechanics

Automated summary

Machine learning techniques provide a new and exciting approach in the field of drug discovery, potentially surpassing traditional molecular mechanics modeling methods. This review article advocates for the combination of both techniques to achieve the most efficient implementation in the coming years. It discusses pure machine learning approaches and recent developments in integrating machine learning with molecular mechanics methods, as well as presents a real industrial prospective study showcasing the successful combination of Monte Carlo molecular mechanics simulations with machine learning for drug-target affinity predictions.