Conformational analysis from crystallographic data using conceptual clustering

The rapid growth of crystallographic databases has created a demand for novel and efficient techniques for the analysis of molecular conformations, in order to derive new concepts and rules and to generate useful classifications of the available data. This paper presents a conceptual clustering approach, termed IMEM (image memory), which discovers the conformational diversity present in a dataset of crystal structures. In contrast to numerical clustering methods, IMEM views a molecular structure as comprising qualitative relationships among its parts, i.e. the structure is viewed as a molecular scene. In addition, IMEM does not require the user to have any a priori knowledge of an expected number of conformational classes within a given dataset. The IMEM approach is applied to several datasets derived from the Cambridge Structural Database and, in all cases, chemically correct and sensible conformational classifications were discovered. This is confirmed by a rigorous comparison of IMEM results with published conformational data obtained by energy-minimization and numerical clustering methods. Conformational analysis tools have an important part to play in the conversion of raw molecular databases to knowledge bases.