FSSP to SCOP and CATH (F2CS) Prediction Server

Summary: The F2CS server provides access to the software, F2CS2.00, that implements an automated prediction method of SCOP and CATH classifications of proteins, based on their FSSP Z-scores (Getz et al., 2002), Availability: Free, at http://www.weizmann.ac.il/physics/complex/compphys/f2cs/. Contact: [email protected] Supplementary information: The site contains links to additional figures and tables. Since during evolution protein structures are much more conserved than sequences and even functions (?), proteins are usually classified first by their structural similarity. Newly solved structures of proteins are regularly stored in the Protein Data Bank (PDB) (?). Many research groups study the diversity of protein structures and maintain web-accessible hierarchical classifications of them. Three widely used databases are FSSP (?), CATH (?) and SCOP (?); although each has its own way to compare and classify proteins, the resulting classification schemes are, largely, consistent with each other (?; ?; ?). The major difference between these three classification schemes, relevant to this work, is their degree of automation. FSSP is based on a fully automated structure comparison algorithm, DALI (?; ?), that calculates a structural similarity measure (represented in terms of Z-scores) between pairs of structures of protein chains taken from the PDB. FSSP first selects a subset of representative structures from the PDB and then applies the DALI algorithm to calculate the Z scores for all pairs of representatives. Next, they calculate the Z scores between each representative and the PDB structures it represents. Being fully automated, FSSP can be updated fairly often. FSSP was recently extended by a new database, called Dali (?), which contains all-against-all Z-scores between chains and domains of a larger representative set, PDB90 (?), in which no two chains are more than 90% sequence identical. In contrast, CATH and SCOP use manual classification at certain levels of their hierarchy, which slows down the classification process and makes it more subjective and error-prone. CATH arranges protein domains in a four-level hierarchy according to their Class (secondary structure composition), Architecture (shape formed by the secondary structures), Topology (connectivity order of the secondary structures) and Homologous superfamily (structural and functional similarity). Classification of Architecture is done by visual inspection; hence CATH is partially manual. The top level (Class) of the SCOP database also describes the secondary structure content of a protein domain. The next level (Fold) groups together structurally similar domains. The lower two levels (superfamily and family) describe near and distant evolutionary relationships (?). ”Fold” largely corresponds to CATH’s topology level (?). SCOP is constructed manually, based on visual examination and comparison of structures, sequences and functions. We present here a web-based server, available at http://www.weizmann.ac.il/physics/complex/compphys/f2cs/, whose aim is to predict, without human intervention, using a protein’s FSSP (or DALI) Z-scores, it’s full SCOP and CATH classifications. This can help classify proteins of known structure that were not yet processed by SCOP or CATH, and call attention to yet unseen structural classes. If a protein appears in FSSP, the server returns our prediction. If it is not in FSSP, the user can submit the new structure to the DALI server, insert the resulting Z-scores into our server and obtain its predicted classification. In both cases F2CS outputs a table showing the prediction, along with its confidence level. THE SERVER The current predictions are based on the latest versions of the databases; FSSP (Jun 16, 2002 update), combined with the Dali database (preliminary version, May 2003); CATH version 2.5 (Jul 2003) and SCOP 1.63 release (May 2003). The FSSP database contains 27182 chains, 2860 out of which are representatives. We superimposed on