How and Why Does Knowledge-Based Biased Docking Improve Molecular Docking Performance?

Since 2012, with the publication of our foundational work "Solvent structure improves docking prediction in lectin-carbohydrate complexes," our group has been devoted to the study of protein-ligand interactions using molecular simulation tools. Over the past decade, we have shown that protein-solvent interactions, particularly when simulated in mixed solvents containing probes such as ethanol, phenol, and isopropanol, often mimic the interactions observed in experimental protein-ligand complexes. This knowledge can be used to improve docking performance by guiding pose prediction and scoring. We termed this strategy-biased docking. Over the years, we demonstrated its applicability to pose prediction, virtual screening (VS), protein-protein docking, and metalloprotein docking. In this short review, we summarize our results and contextualize them within the broader literature, offering a concise description of how to implement the biased docking strategy using current docking software. We also explore the physicochemical rationale behind its effectiveness and discuss how this knowledge can inform emerging Machine Learning and AI-based methodologies.