Assigning protein functions by comparative genome analysis: Protein phylogenetic profiles

Abstract

Determining protein functions from genomic sequences is a central goal of bioinformatics. We present a method based on the assumption that proteins that function together in a pathway or structural complex are likely to evolve in a correlated fashion. During evolution, all such functionally linked proteins tend to be either preserved or eliminated in a new species. We describe this property of correlated evolution by characterizing each protein by its phylogenetic profile, a string that encodes the presence or absence of a protein in every known genome. We show that proteins having matching or similar profiles strongly tend to be functionally linked. This method of phylogenetic profiling allows us to predict the function of uncharacterized proteins.The fully sequenced genomes of numerous organisms offer large amounts of information about cellular biology (see the genomes listed at the web site of The Institute for Genome Research: www.tigr.org). It is a central challenge of bioinformatics to use this information in discovering the function of proteins. Functional assignments of genes come primarily from biochemical experimentation, which can be extended by matching recently sequenced proteins to those that have already been characterized (1). For the exceptionally well studied genome of Escherichia coli (2), these and related techniques (3, 4) have lead to tentative functional assignments of slightly more than half of its proteins (5). The problem of assigning functions to the remaining proteins is addressed here.Our computational method detects proteins that participate in a common structural complex or metabolic pathway. Proteins within these groups are defined as functionally linked. The underlying hypothesis is that functionally linked proteins evolve in a correlated fashion, and, therefore, they have homologs in the same subset of organisms. For instance, we expect to find flagellar proteins in bacteria that possess flagella but not in other organisms. In short, we show that if two proteins have homologs in the same subset of fully sequenced organisms, they are likely to be functionally linked. We exploit this property systematically to map links between all the proteins coded by a genome. In general, pairs of functionally linked proteins have no amino acid sequence similarity with each other and, therefore, cannot be linked by conventional sequence-alignment techniques. METHODSTo represent the subset of organisms that contain a homolog, we constructed a phylogenetic profile for each protein. This profile is a string with n entries, each one bit, where n corresponds to the number of genomes (16 in the present article). We indicate the presence of a homolog to a given protein in the nth genome with an entry of unity at the nth position. If no homolog is found, the entry is zero. Proteins are clustered according to the similarity of their phylogenetic profiles. Similar profiles show a correlated pattern of inheritance and, by implication, functional linkage. The method predicts that the functions of u...

Global protein function prediction from protein-protein interaction networks

“…Proteins in gray boxes are unclassified (unknown function) while the others are classified proteins with the functions within the brackets and labeled according to the following criteria: 1-cell growth; 2-budding, cell polarity and filament formation; 3-pheromone response, mating-type determination, sex-specific proteins; 4-cell cycle check point proteins; 5-cytokinesis; 6-rRNA synthesis; 7-tRNA synthesis; 8-transcriptional control; 9-other transcription activities; 10-other pheromone response activities; 11-stress response; and 12-nuclear organization. Given one of these proteins of unknown function if we take as a prediction the function that appears more in the neighbor proteins of known function then we obtain the following classification (from top to bottom) YNL127W (2), YDR200C (3,4,10) and YLR238W (12). Our method, however, considers also the interactions among unclassified proteins.…”

“…Our method, however, considers also the interactions among unclassified proteins. If we iterate once more the majority rule by taking into account the interactions between the three unclassified proteins, we obtain the following classification YNL127W (2,4), YDR200C (3,4,10) and YLR238W (12). In this way we find another possible function for YNL127W.…”

“…In this context, the search for reliable methods for proteins' function assignment is of uttermost importance. Previous approaches to deduce the unknown function of a class of proteins have exploited sequence similarities or clustering of co-regulated genes [2,3], phylogenetic profiles [4], protein-protein interactions [5,6,7,8], and protein complexes [9,10]. We propose to assign functional classes to proteins from their network of physical interactions, by minimizing the number of interacting proteins with different categories.…”

Assessment of prediction accuracy of protein function from protein–protein interaction data

“…In addition to the standard homology searching, several computational methods for functional prediction of proteins have been proposed. They include the domain fusion method (Marcotte et al, 1999a) and the phylogenetic profile method (Pellegrini et al, 1999), both of which are based on the comparative genomics. Although such methods will give us interesting clues, their reliability is not satisfactory high.…”

Pharmacoproteomics in drug development

“…Two distinct examples include the correlation of mRNA expression profiles to function, as in the characterization of pharmacological perturbations by identifying a novel target of dyclonine, 78 and through in silico predictions of structural or pathway interactions from whole genome comparative analyses. 79 von Mering et al 80 evaluated the roughly 80 000 interactions among yeast proteins currently available in various databases by comparing data from yeast two-hybrid systems, mass spectrometric protein complex purification techniques, mRNA expression profiles, genetic interaction data, as well as in silico interaction predictions from gene fusion, neighborhood, and co-occurrence analysis. Surprisingly, only B2400 of these interactions are supported by more than one method.…”