From simulation data to conformational ensembles: Structure and dynamics-based methods

Statistical methods for analyzing large data sets of molecular configurations within the chemical concept of molecular conformations are described. The strategies are based on dependencies between configurations of a molecular ensemble; the article concentrates on dependencies induced by a) correlations between the molecular degrees of freedom, b) geometrical similarities of configurations, and c) dynamical relations between subsets of configurations. The statistical technique realizing aspect a) is based on an approach suggested by AMADEI ET AL. (Proteins, 17 (1993)). It allows to identify essential degrees of freedom of a molecular system and is extended in order to deter mine single configurations as representatives for the crucial features related to these essential degrees of freedom. Aspects b) and c) are based on statistical cluster methods. They lead to a decomposition of the available simulation data into conformational ensembles or subsets with the property that all configurations in one of these subsets share a common chemical property. In contrast to the restriction to single representative conformations, conformational ensembles include infor mation about, e.g., structural flexibility or dynamical connectivity. The conceptual similarities and differences of the three approaches are discussed in detail and are illustrated by application to simulation data originating from a hybrid Monte Carlo sampling of a triribonucleotide. K e y words , conformational ensemble, cluster method, structural and dynamical similarity, representative, conformation, essential degrees of freedom, transition states, cluster analysis, feature extraction. M a t h e m a t i c s subjec t classification. 6 5 U 0 5 , 62H30 .