Mapping of proteinase active sites by projection of surface-derived correlation vectors

A method for the mapping of surface cavities in proteins is presented. Cavities were defined by a grid-based accessibility measure, where a protein atom was considered to form part of the cavity surface if its accessibility value was below a critical threshold. Surfaces were calculated by the Connolly algorithm. To each surface point, one of five generalized atom types and an accessibility value were assigned. Based on this surface description, topological correlation vectors were generated and projected onto the plane by means of linear and nonlinear mapping techniques. The method was successfully applied to represent relationships among a diverse set of aspartic, serine, and metalloproteinases.