PIRSF: family classi®cation system at the Protein Information Resource

The Protein Information Resource (PIR) is an integrated public resource of protein informatics. To facilitate the sensible propagation and standardization of protein annotation and the systematic detection of annotation errors, PIR has extended its superfamily concept and developed the SuperFamily (PIRSF) classi®cation system. Based on the evolutionary relationships of whole proteins, this classi®cation system allows annotation of both speci®c biological and generic biochemical functions. The system adopts a network structure for protein classi®cation from superfamily to subfamily levels. Protein family members are homologous (sharing common ancestry) and homeomorphic (sharing full-length sequence similarity with common domain architecture). The PIRSF database consists of two data sets, preliminary clusters and curated families. The curated families include family name, protein membership, parent±child relationship, domain architecture, and optional description and bibliography. PIRSF is accessible from the website at http://pir.georgetown.edu/pirsf/ for report retrieval and sequence classi®cation. The report presents family annotation, membership statistics, cross-references to other databases, graphical display of domain architecture, and links to multiple sequence alignments and phylogenetic trees for curated families. PIRSF can be utilized to analyze phylogenetic pro®les, to reveal functional convergence and divergence, and to identify interesting relationships between homeomorphic families, domains and structural classes. INTRODUCTION The Protein Information Resource (PIR) is an integrated public bioinformatics resource that supports genomic and proteomic research and scienti®c studies. For over three decades, PIR has provided many protein databases and analysis tools freely accessible to the scienti®c community, including the PIR-International Protein Sequence Database (PSD) of functionally annotated protein sequences, which grew out of the Atlas of Protein Sequence and Structure (1) edited by Margaret Dayhoff. PIR has recently joined forces with the European Bioinformatics Institute (EBI) and the Swiss Institute of Bioinformatics (SIB) to establish UniProt (the Universal Protein Knowledgebase) (2), the central resource of protein sequence and function, by unifying the database activities of PIR-PSD, Swiss-Prot and TrEMBL. In addition, we have implemented the new PIRSF (SuperFamily) classi®cation system, which is described below. We have also enhanced iProClass (3), an integrated database of protein family, function and structure information with executive summaries and cross-references to over 50 molecular databases; maintained PIR-NREF (4), a non-redundant reference database; and improved the PIR website for scienti®c inquiry and system dissemination. PIRSF SYSTEM DEFINITION The PIR superfamily/family concept (5), the original classi®cation based on sequence similarity, has been used as a guiding principle to provide comprehensive and non-overlapping clustering of PIR protein sequences into a hierarchical order to re ̄ect their evolutionary relationships (6). To facilitate the sensible propagation and standardization of protein annotation and the systematic detection of annotation errors as part of the UniProt project, PIR has extended its hierarchical superfamily concept and developed the PIRSF system, a `network classi®cation system based on the evolutionary relationships of whole proteins'. Classi®cation based on whole proteins, rather than on the component domains, *To whom correspondence should be addressed. Tel: +1 202 687 2121; Fax: +1 202 687 1662; Email: pirmail@georgetown.edu D112±D114 Nucleic Acids Research, 2004, Vol. 32, Database issue DOI: 10.1093/nar/gkh097 Nucleic Acids Research, Vol. 32, Database issue ã Oxford University Press 2004; all rights reserved allows annotation of both generic biochemical and speci®c biological functions. Furthermore, it permits the classi®cation of proteins without well-de®ned domains. The network classi®cation system accommodates a ̄exible number of levels that re ̄ect varying degrees of sequence conservation. Such structure allows improved protein annotation, more accurate extraction of conserved functional residues and classi®cation of distantly related orphan proteins. The primary level for curation is the homeomorphic family, which consists of proteins that are both homologous (evolved from a common ancestor as inferred by detectable sequence similarity) and homeomorphic (sharing full-length sequence similarity and a common domain architecture). Common domain architecture is indicated by the same type, number and order of core domains. Variation may exist for repeating domains and/or auxiliary domains, which are often mobile and may be easily lost, acquired or functionally replaced during evolution. Above the `homeomorphic family' nodes in the network structure are parent superfamily nodes that connect distantly related families and orphan proteins based on common domains. They may be homeomorphic superfamilies, but are more likely to be domain superfamilies if the common domain regions do not extend over the entire full-length proteins. Below the homeomorphic family nodes are child subfamily nodes, which are homologous and homeomorphic clusters representing functional specialization and/or domain architecture variation within a family. The PIRSF system de®nition and working principles are detailed in the document, A Proposal for the PIRSF Classi®cation System, available from the PIR website. PIRSF DATABASE CREATION AND CURATION The PIRSF database consists of two data sets, preliminary clusters and curated families. Currently, about two-thirds of UniProt sequences are classi®ed into over 32 000 preliminary clusters, including single-member clusters. The preliminary clusters are computationally de®ned using both pairwisebased parameters (% sequence identity, sequence length ratio and overlap length ratio) and cluster-based parameters (% matched members, distance to neighboring clusters and overall domain arrangement). Systematic family curation is being conducted in a two-tier process to improve the quality of automated classi®cation. Over 4500 families containing two or more members have been curated at the `®rst-tier' for membership and domain architecture characteristic of the family. PIRSF has two membership types: regular members for proteins sharing endto-end sequence similarity and associate members for proteins whose lengths deviate from the family length range, including incomplete sequences, alternate splice and initiator variants, and peptides derived from proteolytic processing. A subset of representative regular members is chosen as seed members for generating multiple sequence alignments, phylogenetic trees and Hidden Markov Models (HMMs) of the respective families. The second-tier curation provides additional annotation, including family name, parent±child relationship, family description and bibliography. Several hundred second-tier curated PIRSF families have been integrated into InterPro (7). The incorporation of PIRSF families into InterPro and the implementation of a system to check the validity and integrity of existing families create additional means of ensuring accuracy and consistency in UniProt classi®cation and annotation.