Sampling low free energy protein structures in high dimensional conformational landscapes
Alberto Perez
University of Florida, Department of Chemistry, P.O. Box 117200, Gainesville, FL 32611
A big challenge for computational structural biology is to study the conformational landscapes of macromolecules. When exhaustive sampling is not enough due to frustrated landscapes and kinetic trapping such as protein folding we need new ways of balancing exploration and exploitation. In this scenario is often more important to concentrate the sampling in regions of low free energy rather than exploring the whole landscape. We have developed a method, MELD (Modeling Employing Limited Data), that incorporates ambiguous, sparse and noisy data into simulations. It allows to sample regions that are compatible with some subset of the data with higher probability through Bayesian inference. More importantly, the relative populations between different minima remain the same as in the original framework. In this work we show how to combine unassigned NOE data from NMR with molecular dynamics simulations of proteins to predict protein structures.