A database of globular protein structural domains: clustering of representative family members into similar folds

Abstract

Backgound: A database of globular domains, derived from a non-redundant set of proteins, is useful for the sequence analysis of aligned domains, for structural comparisons, for understanding domain stability and flexibility and for fold recognition procedures. Domains are defined by the program DIAL and classified structurally using the procedure SEA. Results: The DIAL-derived domain database (DDBASE) consists of 436 protein chains involving 695 protein domains. Of these, 206 are α-class, 191 are β-class and 294 α and β class. The domains, 63% from multidomain proteins and 73% less than 150 residues in length, were clustered automatically using both single-link cluster analysis and hierarchical clustering to give a quantitative estimate of similarity in the domain-fold space. Conclusion: Highly populated and well described folds (doubly wound α/β, singly wound α/β barrels, globins α, large Greek-key β and flavin-binding α/β) are recognized at a SEA cut-off score of 0.55 in single-link clustering and at 0.65 in hierarchical clustering, although functionally related families are usually clearly distinguished at more stringent values.