Biased Metropolis Sampling for Rugged Free Energy Landscapes

TL;DR

This paper introduces a biased Metropolis sampling method that uses high-temperature probability densities to enhance the efficiency of peptide simulations in rugged free energy landscapes, reducing autocorrelation times.

Contribution

The authors propose a novel transformation of updating probabilities based on higher temperature densities, improving sampling efficiency in peptide simulations.

Findings

Autocorrelation times are nearly halved using the new method.

The approach benefits both canonical and generalized ensemble simulations.

Performance improvements multiply when combined with generalized ensemble techniques.

Abstract

Metropolis simulations of all-atom models of peptides (i.e. small proteins) are considered. Inspired by the funnel picture of Bryngelson and Wolyness, a transformation of the updating probabilities of the dihedral angles is defined, which uses probability densities from a higher temperature to improve the algorithmic performance at a lower temperature. The method is suitable for canonical as well as for generalized ensemble simulations. A simple approximation to the full transformation is tested at room temperature for Met-Enkephalin in vacuum. Integrated autocorrelation times are found to be reduced by factors close to two and a similar improvement due to generalized ensemble methods enters multiplicatively.