Protein motifs: towards a unified view on databases

Identification of biologically important motifs in protein sequences requires the parallel handling of the structural and biological data of proteins given in sequence databases. We approach this problem with a generalized data model, in which both kinds of data are included in one representation consisting of substructures (entities), relationships, and hierarchical classification schemes containing the semantic definitions of these. A consistent implementation of this model is not attempted because of the lack of standardized semantic classification schemes. However, we use the model to design simple approximate strategies that can identify patterns in current (partially standardized) databases. Application examples to prediction of functional domains from sequence are given.<>