Machine Learning Enhanced Molecular Docking: Advances in Algorithms, Accuracy, and Drug Discovery Efficiency

Molecular docking is one of the core algorithms used in structure-based drug discovery, which makes it possible to compute binding orientations and affinities between small molecule ligand and a biological target. Standard docking algorithms are based on physics based scoring functions and heuristic search strategies, which can find the trade-off between predictive accuracy and computational efficiency difficult. In recent years, the field of machine learning (ML) has developed as a disruptive paradigm that is able to learn complicated, non-linear relationships using large-scale biochemical information. ML-based methods applied in conjunction with molecular docking have greatly improved the quality of algorithmic performance, improved prediction of binding affinity, and speeded up virtual screening pipelines. In this paper, a rigorous and in-depth review of the machine learning-based molecular docking methods is provided and centered on algorithmic developments, accuracy, and efficiency enhancements in pharmaceutical discovery processes. We address supervised, unsupervised, and deep learning methods that are used in pose prediction, optimization of any scoring function, and docking