Mapping the Protein Conformational Landscape with Adaptive Probabilistic Search

Abstract

Structural characterization of the protein native state is often the key to better understanding protein function. The conformations that comprise the native state reside in the lowest-energy basin of a funnel-like energy landscape. Discovering and populating this basin with native conformations in silico demands powerful search algorithms that can navigate high-dimensional conformational spaces. Even short protein chains have many degrees of freedom. Additionally, semi-empirical energy functions employed to guide the conformational search may introduce slight distortions to the true energy landscape, often leading to rough landscapes rich in local minima. A successful search algorithm must balance the competing goals of sampling a diverse representation of the landscape with the need to further populate promising energy minima. We present a novel algorithm which effectively balances the goals of exploration and exploitation through the use of projection layers. A geometric layer keeps track of the structural diversity in the explored conformational space, and an energetic layer determines the relevance of a computed conformation for the native state. The algorithm conducts a probabilistic search of a coarse-grained conformational space, maintaining a representative ensemble of computed conformations. A probabilistic weighting function over the projection layers determines where to guide the search in the conformational space by balancing coverage of conformational space with population of lower-energy levels. We have compiled an extensive list of structurally-diverse proteins on which we apply our algorithm. Our results show that the algorithm efficiently yields native-like coarse-grained conformations of diverse small-to-medium size proteins of alpha, beta, and alpha/beta folds. The conformational ensemble maintained by the algorithm provides a representative map of the conformational landscape of each protein. The lowest-energy conformations in this map capture the native state and reproduce well the known native structures upon further refinement with all-atom energy functions.