AUTOMATED CONFORMATIONAL-ANALYSIS FROM CRYSTALLOGRAPHIC DATA .3. 3-DIMENSIONAL PATTERN-RECOGNITION WITHIN THE CAMBRIDGE STRUCTURAL DATABASE SYSTEM - IMPLEMENTATION AND PRACTICAL EXAMPLES

A unified computational procedure is described for the identification of conformational subgroups for a chemical fragment from crystal structure data. Fragment conformations are defined by N(t) torsion angles for N(f) occurrences of the fragment in the Cambridge Structural Database. Subdivision of this multivariate data set is performed by a choice of clustering algorithms (single-linkage, complete-linkage, Jarvis-Patrick). Both asymmetric and symmetric fragments are handled routinely. The algorithms yield optimum superposition of a given conformation in a single cluster and place discrete clusters into a single asymmetric unit of conformational space. The unified procedure generates graphical and numerical indicators of clustering efficiency: (i) principal-component plots of the optimally superimposed data set, (ii) a simple statistical summary for each cluster, (iii) measures of intracluster shape and size, (iv) details of intercluster separations. Major clusters are ranked in order of decreasing population and the 'most representative fragment' (MRF: the fragment of the data set which is closet to the cluster centroid) is identified in each case. Atomic coordinates for the MRF's may be output for use as conformational alternatives in model building. The complete procedure is successfully applied to the automated conformational analysis of two very different systems, the cyclic 1-azacycloheptane moiety and the acyclic C-17 side chain typical of cholesterol and related steroids.