Year: 2026
Type: preprint
Abstract: Accurate and efficient prediction of binding free energies remains a central challenge in computational chemistry, largely due to the complexity of solvent effects and the trade-off between accuracy and computational cost. Explicit solvent molecular dynamics (MD) simulations provide detailed thermodynamic and kinetic insight, now routinely reaching microsecond timescales, while end-point methods such as MM/GBSA offer high efficiency but are sensitive to model parameters. In particular, the accuracy of MM/GBSA strongly depends on the atomic radii used to define the dielectric boundary within the implicit solvent framework. Here, we introduce an explicit–implicit hybrid workflow for late-stage ligand prioritization that integrates the strengths of both approaches. In this framework, microsecond-scale explicit solvent MD simulations act as a rapid-dissociation filter, flagging weak or unstable binders through spontaneous dissociation within the simulation window. MM/GBSA is then used to rank the remaining candidates. To improve the accuracy of the implicit solvent stage, we employ optimized atomic radii sets (OPTBIND), specifically designed for binding free-energy calculations. On the protein–ligand complexes of Curated32-4 set , a new curated benchmark of 32 complexes including 4 congeneric series, the entropy-free MM/GBSA stage with OPT- BIND 6D achieved Pearson r = 0.82 and Spearman ρ = 0.84, with 100% recovery of the experimental rank ordering across all four congeneric series, while a single-replica consistency test of the rapid-dissociation filter yielded an AUC of 0.88 and retained all 15 sub-micromolar binders. Finally, as a real-world drug design application, we apply the workflow to halogen substitutions in SARS-CoV-2 main protease inhibitors, demonstrating its ability to identify favorable chemical modifications. Notably, the approach prospectively identifies F→Cl and F→I as affinity-enhancing substitutions; the predicted enhancement for the chlorine analog is subsequently confirmed experimentally, while the iodine analog remains a testable prediction, highlighting its predictive power in guiding chemical optimization. Together, these results establish a practical framework for detailed studies of receptor–ligand binding and for late-stage ligand prioritization in drug discovery.
Source: ChemRxiv
Institutions: Virginia Tech, Roanoke College
Language: English
Subfield: Drug Discovery
Domain: Life Sciences
Open Access status: green
OpenAlex ID: W7221555044