Trajectory-Based Parameterization of a Coarse-Grained Forcefield for High-Throughput Protein Simulation

The traditional trade-off in biomolecular simulation between accuracy and computational efficiency is predicated on the assumption that detailed forcefields are typically well-parameterized (i.e. obtaining a significant fraction of possible accuracy). We re-examine this trade-off in the more realistic regime in which parameterization is a greater source of bias than the level of detail in the forcefield. To address parameterization of coarse-grained forcefields, we use the contrastive divergence technique from machine learning to train directly from simulation trajectories on 450 proteins. In our scheme, the computational efficiency of the model enables high accuracy through precise tuning of the Boltzmann ensemble over a large collection of proteins. This method is applied to our recently developed Upside model [1], where the free energy for side chains are rapidly calculated at every time-step, allowing for a smooth energy landscape without steric rattling of the side chains. After our contrastive divergence training, the model is able to fold proteins up to approximately 100 residues de novo on a single core in CPU core-days. Additionally, the improved Upside model is a strong starting point both for investigation of folding dynamics and as an inexpensive Bayesian prior for protein physics that can be integrated with additional experimental or bioinformatic data.