Evidence for the Progressive Improvement of All-Atom Force Fields in Reproducing Local Conformational Preferences of Flexible Peptides

Abstract

AbstractClassical protein force fields are widely used to probe the conformational properties of intrinsically disordered regions, yet their accuracy in reproducing local structural preferences remains uneven. We evaluated seven Amber and CHARMM force fields across three generations using molecular dynamics simulations of glycine-X-glycine tripeptides, with guest residues that span diverse physicochemical properties. Conformational ensembles were compared against distributions of conformations extracted from the crystallographic structures in the Protein Data Bank, and a statistical model derived from NMR observables. Analysis of secondary structure populations and Ramachandran distributions analyzed via Wasserstein distances reveals a clear historical progression. Early models display strong helical bias, intermediate ones approach Protein Data Bank trends, and recent versions shift toward solution-like ensembles dominated by polyproline II structure. None of the force fields fully captures the experimental distributions, although recent models show marked improvement over earlier generations. The remaining discrepancies point to specific aspects of local structure that still require tuning, while the overall progress underscores a steady trajectory toward more reliable descriptions of disordered peptides.