Advances in Molecular Docking Methodologies

Computer-aided drug design (CADD) is undergoing a fundamental paradigm shift driven by the transition from classical biophysical methods to deep learning architectures and generative artificial intelligence. This review analyzes the evolution of molecular docking algorithms. We examine traditional programs (AutoDock Vina, Glide, GOLD) based on stochastic conformational search and empirical scoring functions, which retain the status of gold standard due to the high physical validity of the generated predictions. Software solutions for high-throughput virtual screening, such as distributed pipelines like EasyDock and graphical interfaces like EasyDockVina, are analyzed. Particular attention is paid to the latest generative AI models (DiffDock, GNINA, AlphaFold 3, DynamicBind, FABFlex), which address the computational challenges of blind docking and macromolecular receptor flexibility. We assess the systemic crisis of neural network generalization ability identified in independent benchmarks (PoseBusters, Bento, NextTopDocker) and substantiate the need to integrate the laws of molecular physics into the latent spaces of models. We conclude that the formation of hybrid pipelines, combining the speed of AI with the rigor of classical mechanics, is a necessary development.