Hamiltonian Replica Exchange Simulations to Enhance Sampling for Protein Folding

Abstract

In order to enhance sampling in biomolecular simulation many efforts have focused on reducing the effective degrees of freedom and employing coarse-grained models. Structure-based or Go-models for protein folding are based on the energy landscape theory and the principle of minimal frustration have demonstrated good agreement with experimental measurements and are computationally sufficiently tractable to allow simulating large-scale structural transitions and provide sufficient sampling even of large-scale conformational transitions. Linking these minimal or coarse-grained models to physically motivated empirical force fields would enhance understanding of the underlying physical/chemical interactions.Here, we present a novel approach combining an efficient coarse-grained native structure-based model with a physics based all-atom model. This approach targets the sampling problem that has plagued traditional simulation methods for decades while providing an energetically accurate description of conformational transitions. Our approach is based on Hamiltonian exchange, a variant of the Replica-exchange concept, by coupling the Hamiltonians that operate not on different temperatures, but on different levels of representation. Each level is occupied by a combination of the two mixed Hamiltonians Htotal=λH1+(1-λ)H2. During simulation, one permits exchanges between neighboring levels. We observe frequent transitions between neighboring levels and an enhanced sampling efficiency for model proteins.