Predictive Modeling of Protein Folding Dynamics

Abstract

Protein folding dynamics describe the time-dependent process by which proteins adopt their native three-dimensional structures, and predictive modelling of folding trajectories, intermediates, and kinetics complements static structure prediction by revealing misfolding pathways, allosteric motions, and conformational ensembles, yet adoption varies across methodological categories. We evaluated 216 predictive folding dynamics programmes across centres in France, Germany, and Italy between 2018 and 2022, spanning five model categories: classical molecular dynamics folding simulations, Markov state models of folding intermediates, neural network enhanced sampling, coarse-grained folding pathway models, and deep learning conformational ensemble generators. A Protein Folding Dynamics Index (PFDI) was constructed from five sub-scores -- trajectory accuracy versus experimental observables, kinetic prediction fidelity, conformational ensemble completeness, computational efficiency, and applicability to clinically relevant systems -- with weights from regression against sustained adoption. PFDI correlated with adoption at r = +0.84 and discriminated adopted from non-adopted methods with an AUC of 0.882. Deep learning ensemble generators scored highest (mean PFDI 0.824), while coarse-grained pathway models trailed at 0.600. Only 35.5 percent exceeded the 0.75 threshold. Trajectory accuracy carried the largest weight (beta = +0.278), followed by conformational ensemble completeness (beta = +0.230).