One classical force field accurate for folded proteins, disordered proteins and weak protein–protein association

• Compaction problem. Earlier generations of force fields produce overly compact disordered states and overly stable protein–protein association (Piana et al., J. Phys. Chem. B 2015, on water dispersion; Lindorff-Larsen et al., PLoS ONE 2012, on folded proteins).
• Compromise force fields. a99SB-disp (Robustelli, Piana and Shaw, PNAS 2018) was parametrised with a modified TIP4P-D water and benchmarked on 21 systems with more than 9,000 experimental data points covering folded proteins, peptides and disordered proteins; the authors report that none of the force fields they tested captured all of folded-protein stability, disordered-protein dimensions and secondary-structure propensities at once before their revision. CHARMM36m (Huang et al., Nat. Methods 2017) improves disordered ensembles, but the Robustelli et al. benchmark reports that it does not fully remove over-compaction of disordered proteins.
• Remaining issues. Phosphorylated and charged disordered proteins, membrane proteins, nucleic-acid/protein condensates and protein–protein affinities remain error-prone in all fixed-charge force fields; polarisable force fields (AMOEBA, Drude) are costlier and not yet shown to resolve the trade-off.
• ML-based. Machine-learned potentials for large proteins are an active direction but not yet validated on this benchmark class.

Predictive simulation of disordered regions, condensates and signalling interactions (a large fraction of the human proteome is disordered), and of drug binding in flexible targets; fewer ad hoc force-field choices in structural biology.

A published force field meeting the four targets on a pre-registered benchmark set (at least 30 disordered, 20 folded and 20 association systems), reproduced by an independent group using a different MD engine.