"Can Coarse-Grained Simulations Reach Chemical Accuracy? Revisiting the Sampling–Accuracy Trade-off"

Despite substantial progress in computational chemistry, reconciling the accuracy of all-atom molecular dynamics (AA-MD) with the efficiency required for high-throughput binding free energy predictions remains a central challenge in structure-based drug discovery. Recently, we proposed address this gap using coarse-grained funnel metadynamics (CG-FMD) [1] based on the Martini 3 force field [2,3]. This framework combines the computational efficiency of coarse-grained representations with enhanced sampling techniques capable of modelling full ligand binding and unbinding pathways, including access to deeply buried binding sites. Results on model systems [1] showed very good agreement with experimental references. Furthermore, thanks to the extensive sampling efficiently achievable with limited HPC resource allocation, the statistical uncertainty of CG-FMD estimations was much reduced compared to the AA-FMD counterpart. Indeed, we suggest that the improved sampling capabilities of CG simulations can partially compensate the simplified representation of the protein-ligand complex. We will show preliminary results [4] of this methodology applied to the exploration of the tubulin αβ heterodimer, a complex multisite protein of strategic pharmaceutical relevance. In particular, we will highlight how CG-FMD has made it possible to obtain binding free energy predictions comparable to experimental reference for pharmaceutical scaffolds binding to the deeply buried colchicinoids site. Overall, we suggest that CG-FMD can become a valuable and efficient physics-based approach for the investigation of protein-ligand interaction on complex biosystems, with high scalability of system size. References [1] Grazzi A, et al. JCTC 2026, doi: 10.1021/acs.jctc.5c01785 [2] Souza, P. C. T., et al. Nature Methods 2021, doi: 10.1038/s41592-021-01098-3 [3] Souza, P. C. T., et al. Nature Communications 2020, doi: 10.1038/s41467-020-17437-5 [4] Grazzi, A. et al. bioRxiv 2026, doi: 10.64898/2026.02.24.707696