Network Analysis and Kinetic Clustering Methods Applied to Molecular Dynamics Simulations of Proteins in Complex Environments

Abstract

Molecular dynamics (MD) simulations are increasingly used to probe the structural dynamics of proteins in response to perturbations in their environment. Characterization of MD trajectories within the simulation time-scale of tens to hundreds of microseconds is hindered by the determination of relevant reaction coordinates that characterize macrostates and transition barriers among them. Over recent years, kinetic clustering methods have been successfully applied to analyze MD trajectories allowing to analyze and describe complex free-energy landscapes of proteins. These methods rely on the identification of a set of relevant features describing these landscapes; however, exhaustive feature identification is limited by the amount of memory available in standard computer clusters. Network analysis and community detection methods provide an alternative to generate coarse-grained features by identifying communities of protein residues that are grouped together according to their time-dependent association. Such communities can be used for coarse-graining the representation of relatively large proteins (several hundreds of residues) to a few communities, over a time-scale of several microseconds. Here, we implement the application of community detection methods to MD trajectories of proteins ranging from tens to several hundreds of residues solvated in explicit water and/or embedded in lipid bilayers. Kinetic clustering methods are applied on both the coarse-grained representation generated by network analysis and the all-atom representation and used to analyze the free-energy landscape of a diverse set of proteins.