A library-based Monte Carlo technique enables rapid equilibrium sampling of a protein model with atomistic components

TL;DR

This paper introduces a library-based Monte Carlo method that enables rapid and statistically rigorous sampling of fully flexible atomistic protein models, significantly improving simulation efficiency.

Contribution

The authors develop a novel Monte Carlo technique using pre-generated libraries of atomistic components to enhance protein simulation speed without losing accuracy.

Findings

Can generate at least 30 independent configurations per protein in about a month on a single CPU.

Minimal additional cost for incorporating residue-specific interactions.

Effective for small proteins with atomistic backbone components.

Abstract

There is significant interest in rapid protein simulations because of the time-scale limitations of all-atom methods. Exploiting the low cost and great availability of computer memory, we report a Monte Carlo technique for incorporating fully flexible atomistic protein components (e.g., peptide planes) into protein models without compromising sampling speed or statistical rigor. Building on existing approximate methods (e.g., Rosetta), the technique uses pre-generated statistical libraries of all-atom components which are swapped with the corresponding protein components during a simulation. The simple model we study consists of the three all-atom backbone residues -- Ala, Gly, and Pro -- with structure-based (Go-like) interactions. For the five different proteins considered in this study, LBMC can generate at least 30 statistically independent configurations in about a month of single CPU time. Minimal additional cost is required to add residue-specific interactions.