AbstractBackgroundResultsPrinciple of the PRODISTIN method and classification of the yeast proteomePRODISTIN clustering depends neither on sequence similarity nor on biochemical functionClassification of the S. cerevisiae proteome: integrated analysis of cellular processes and their cross-talkTable 1 Functional predictions and their biological relevanceTable 2 Statistical evaluations of PRODISTIN clustersDiscussionProtein-protein interactions as good indicators of protein cellular functionComparison of the PRODISTIN method with recent functional prediction methodsConclusionsMaterials and methodsProtein-protein interaction data setsClassification methodSequence alignments and analysisSubtree robustness measurementAnnotation sources and functional tree visualizationAdditional data filesAcknowledgementsReferencesGenome Biology 2003, 5:R6comment reviews reports deposited research refereed research interactions informationOpen Access2003Brunet al.Volume 5, Issue 1, Article R6MethodFunctional classification of proteins for the prediction of cellular function from a protein-protein interaction networkChristine Brun*, François Chevenet¤†, David Martin¤*, Jérôme Wojcik‡, Alain Guénoche§ and Bernard Jacq*Addresses: *Laboratoire de Génétique et Physiologie du Développement, CNRS UMR6545, Parc Scientifique et Technologique de Luminy, Case 907, 13288 Marseille Cedex 9, France. †Centre d'Etude sur le Polymorphisme des Micro-organismes, CNRS/IRD UMR 9926, 911 avenue Agropolis, BP 6450, 34394 Montpellier Cedex 5, France. ‡Hybrigenics SA, 3/5 impasse Reille, 75014 Paris, France. §Institut de Mathématiques de Luminy, CNRS UPR9016, Parc Scientifique et Technologique de Luminy, Case 907, 13288 Marseille Cedex 9, France. ¤ These authors contributed equally to this work.Correspondence: Bernard Jacq. E-mail: [email protected]© 2003 Brun et al.; licensee BioMed Central Ltd. This is an Open Access article: verbatim copying and redistribution of this article are permitted in all media for any purpose, provided this notice is preserved along with the article's original URL.Functional classification of proteins for the prediction of cellular function from a protein-protein interaction networkWe here describe PRODISTIN, a new computational method allowing the functional clustering of proteins on the basis of protein-protein interaction data. This method, assessed biologically and statistically, enabled us to classify 11% of the Saccharomyces cerevisiae proteome into several groups, the majority of which contained proteins involved in the same biological process(es), and to predict a cellular function for many otherwise uncharacterized proteins.AbstractWe here describe PRODISTIN, a new computational method allowing the functional clustering ofproteins on the basis of protein-protein interaction data. This method, assessed biologically andstatistically, enabled us to classify 11% of the Saccharomyces cerevisiae proteome into several groups,the majority of which contained proteins involved in the same biological process(es), and to predicta cellular function for many otherwise uncharacterized proteins.BackgroundComplete genome sequencing makes available a largenumber of coding protein sequences for which we have littleor no functional information. In fact, the function of 30-35%of encoded proteins per completely sequenced genomeremains unknown [1]. To decipher the functions of these pro-teins and, more broadly, to propose functional relationshipsamong proteins, new computational methods relying upongenome organization have been developed. The RosettaStone method proposes that two proteins in a given proteomeare functionally linked when they exist as a single fusedpolypeptide in another proteome [2,3]. The chromosomalproximity method suggests that genes repeatedly found asneighbors on chromosomes in different organisms mayencode functionally related proteins [4-6]. Finally, the phylo-genetic co-inheritance of proteins in several different pro-teomes may indicate their functional link [7]. Although thesemethods and combinations thereof [8] successfully predictthe function of certain proteins, they suffer from several lim-itations: they are more informative when applied to com-pletely sequenced genomes; they are generally moreappropriate for prokaryotic genome organization; and theprinciples underlying some of them are only valid for a smallnumber of proteins.Molecular interactions are essential actors for all biologicalprocesses. Large-scale studies of protein-protein interactionshave been carried out in several organisms to establish inter-action maps and to decipher protein function [9-16]. Theselarge intricate networks now need to be analyzed in detail toextract information related to protein function and to rela-tionships linking cellular processes. Various methods of bio-logical network analysis have been proposed so far. They may,for instance, allow identification of functional modules afternetwork clustering [17], or the assignment of function to pro-teins of unknown function on the basis of the functionalPublished: 15 December 2003Genome Biology 2003, 5:R6Received: 25 June 2003Revised: 6 October 2003Accepted: 14 November 2003The electronic version of this article is the complete one and can be found online at http://genomebiology.com/2003/5/1/R6R6.2 Genome Biology 2003, Volume 5, Issue 1, Article R6 Brun et al. http://genomebiology.com/2003/5/1/R6Genome Biology 2003, 5:R6annotation of their neighbors [18]. Another way to analyzethe interaction network is to compare proteins functionally atthe cellular level. This approach would represent a usefulcomplement to sequence-comparison methods, whichaddress function at the molecular level. With this in mind, wepropose a new bioinformatics method allowing a functionalclassification of the proteins according to the identity of theirinteracting partners.The method, named PRODISTIN for protein distance basedon interactions, was applied to the yeast interactome and sta-tistically evaluated for robustness using several independentcriteria. The analysis of the results obtained demonstratedthat proteins are grouped according to their cellular ratherthan molecular function; proteins involved in the samemolecular complex(es), pathway(s) or cellular process(es) areclustered; a sound prediction of cellular function for theuncharacterized proteins is possible. The biological relevanceof the obtained predictions is discussed with respect
View Full Document
Unlocking...