A Multiscale Funnel Metadynamics Framework for Protein–Ligand Binding Free Energy Landscapes

Accurate prediction of protein–ligand binding free energies remains a central challenge in computational biophysics, particularly when balancing atomistic accuracy with the extensive sampling required to characterize complex binding landscapes. Funnel metadynamics (FMD) has emerged as a powerful enhanced-sampling approach for reconstructing ligand binding/unbinding pathways and estimating absolute binding free energies using funnel-shaped restraining potentials [1]. Building upon recent developments in coarse-grained funnel metadynamics (CG-FMD) based on the Martini 3 force field, we present a dual-resolution strategy that combines all-atom (AA) and coarse-grained (CG) representations to investigate protein–ligand recognition processes. In this work [2], we compare AA-FMD and CG-FMD free energy calculations across representative protein–ligand systems and demonstrate that the two resolutions produce remarkably consistent free energy surfaces (FESs), despite the reduced complexity of the CG representation. In particular, the qualitative topology of the binding pathways, and the estimated binding affinities are preserved between AA and CG simulations. These results suggest that CG-FMD can retain the essential thermodynamic determinants of ligand recognition while dramatically reducing computational cost and enabling significantly longer effective sampling times. The agreement between AA and CG FESs suggests that coarse-grained enhanced sampling can be exploited not only as a rapid screening tool, but also as a physically meaningful framework for guiding higher-resolution simulations. As a challenging application, we are now focusing on tubulin–ligand recognition [3], where the rugged conformational landscape and the presence of multiple metastable states complicate convergence in atomistic simulations. Here, we explore a multiscale workflow in which CG-FMD is first employed to obtain a preliminary reconstruction of the binding free energy landscape and to identify the dominant ligand binding/unbinding pathways. These CG-derived pathways are subsequently used to define physics-informed collective variables (CVs) for AA-FMD simulations, thereby improving the efficiency of enhanced sampling in complex systems. Future results will clarify whether this improved simulation protocol can capture energetic barriers along the path, with important implications for kinetics and pharmacological insights. Overall, our ongoing simulations suggest that the integration of CG-FMD and AA-FMD within a unified multiscale framework may provide a computationally efficient strategy for protein–ligand binding studies, with the potential to enable accurate free energy estimation and the development of physics-informed collective variables for challenging biomolecular targets.