Prediction of evolutionarily conserved interologs in Mus musculus
© Yellaboina et al; licensee BioMed Central Ltd. 2008
Received: 05 May 2008
Accepted: 08 October 2008
Published: 08 October 2008
Identification of protein-protein interactions is an important first step to understand living systems. High-throughput experimental approaches have accumulated large amount of information on protein-protein interactions in human and other model organisms. Such interaction information has been successfully transferred to other species, in which the experimental data are limited. However, the annotation transfer method could yield false positive interologs due to the lack of conservation of interactions when applied to phylogenetically distant organisms.
To address this issue, we used phylogenetic profile method to filter false positives in interologs based on the notion that evolutionary conserved interactions show similar patterns of occurrence along the genomes. The approach was applied to Mus musculus, in which the experimentally identified interactions are limited. We first inferred the protein-protein interactions in Mus musculus by using two approaches: i) identifying mouse orthologs of interacting proteins (interologs) based on the experimental protein-protein interaction data from other organisms; and ii) analyzing frequency of mouse ortholog co-occurrence in predicted operons of bacteria. We then filtered possible false-positives in the predicted interactions using the phylogenetic profiles. We found that this filtering method significantly increased the frequency of interacting protein-pairs coexpressed in the same cells/tissues in gene expression omnibus (GEO) database as well as the frequency of interacting protein-pairs shared the similar Gene Ontology (GO) terms for biological processes and cellular localizations. The data supports the notion that phylogenetic profile helps to reduce the number of false positives in interologs.
We have developed protein-protein interaction database in mouse, which contains 41109 interologs. We have also developed a web interface to facilitate the use of database http://lgsun.grc.nia.nih.gov/mppi/.
Many functions in living organisms are determined by interactions among proteins in cells. Identifying these interactions is an important first step in systems level understanding of various developmental, physiological, and disease processes. High-throughput experimental approaches such as yeast two-hybrid system and tandem affinity purification coupled with mass spectrometry have been carried out to map protein-protein interactions in model organisms [1–6]. These experimental data have been curated to produce protein-protein interaction databases such as BIOGRID , INTACT , MINT , DIP , and Reactome . Computational methods have also been developed to transfer the interaction annotation from one organism to another through identifying orthologs by comparative genomics methodology [12, 13].
In addition to the experimental approaches mentioned above, a number of algorithms have been developed to predict protein-protein interactions by computationally analyzing completely sequenced genomes. Some of these algorithms identify the interactions between proteins on the basis of chromosomal proximity of two genes. These methods rely on the notion that genes encoding functionally interacting proteins show conserved gene neighborhood and are often localized in gene clusters or operons in the bacterial genomes [14–17]. Special case of chromosomal proximity is a gene-fusion, where the fusion between two genes in another genome is usually a strong indication for a physical interaction between the proteins encoded therein . Regardless of the proximity in the chromosome, being encoded in the same genome and their co-evolution can be a prerequisite for functional interaction. One such approach is a phylogenetic profile method that identifies interactions by using the pattern of occurrence of genes or protein domains in genomes of different species [19–21]. Other coevolution methods are an in-silico two hybrid system and mirror tree method, which detects interactions between the proteins on the basis of correlated mutations and similarity of phylogenetic trees, respectively .
A potential problem in predicting protein-protein interactions using such an interolog-based method is that it may generate false positive interactions, because of false positives in the original high-throughput interaction data [23, 16] and false positive interologs due to the lack of evolutionary conservation of interactions when applied to phylogenetically distant organisms . Topology of the network (quasi clique score) has been used to filter out the false positives interologs on the basis of notion that the highly interconnected proteins are likely to be evolutionary conserved . For accurate transfer of interactions to orthologs, HomoMINT uses domain matching algorithm to filter the false positives in orthologs . Because the interacting proteins are likely to show similar functions, functional similarity of gene ontology terms has been used to reduce the false positives in high-throughput protein-protein interaction data [27, 28].
In the present work, we inferred the interactions between Mus musculus proteins, if their orthologs are known to be interacting in other species or part of predicted operons in bacteria. We anticipated the presence of false positives in predicted interactions due to the lack of evolutionary conservation of interactions. To reduce the false positives, we have used the phylogenetic profiles of interacting proteins and filtered out unlikely interactions.
2 Results and discussion
2.1 Transferring experiment-based interologs of model organisms to Mus musculus
Distribution of interologs in mouse before and after filtering with phylogenetic profiles
Non-redundant interactions identified experimentally in each species
Interologs transferred to mouse based on orthologous relationship
Interologs remained after filtering by phylogenetic profiles
Fraction of filtered interologs (%)
Escherichia coli K12
We transferred the interactions from each of the reference organisms to Mus musculus on the basis of orthology relationship predicted as the best hit by bi-directional BLASTp  searches against all proteins using an 10-4 as a cut-off e-value. As for the redundant interologs in mouse, an interolog from evolutionarily more closely related species was selected. Table 1 shows the combined non-redundant set of original interactions for each reference organism and Mus musculus. Homo sapiens contributed the greatest number of interologs to Mus musculus, followed by Saccharomyces cerevisiae. After removing redundancy, we obtained a total 55913 non-redundant interologs in Mus musculus. The final number of interactions consisting of interologs and interactions identified experimentally in Mus musculus was 58352 [See Additional file 1].
2.2 Predicting interactions based on the co-occurrence of mouse orthologs in predicted bacterial operons
We used the support vector machine (SVM) that was trained on intergenic distances to predict the operons in 186 species of bacteria, as described previously  [See Additional file 2]. When two mouse orthologs were found at least in one predicted operon, we considered these two mouse proteins interacting either functionally or physically. Using this method, we identified 7870 interactions between 2054 proteins in Mus musculus [See Additional File 3]. In general, the reliability of the interaction decreases as ortholog frequency of co-occurrence in predicted operons decreases. To make this point clear, we sorted the interactions by the decrease in frequency of co-occurrence in predicted operons and showed it in the column 5 in Additional file 3.
We combined 58352 experiment-based interologs and 7870 operon-based interactions, removed redundancy, and obtained 65515 protein-protein interactions in Mus musculus. Relatively low overlap (707 common interactions) between experiment-based interologs and operon-based interologs may possibly be due to the scarcity of the known interactions.
2.3 Filtering false positives using phylogenetic profiles
The interactions transferred from both experimentally identified interactions from model organisms (section 2.1) and predicted operons of bacteria (section 2.2.) may include false positives due to false positives in original interactions obtained experimentally or the lack of evolutionary conservation of interactions. We, therefore, used a phylogenetic profile of 26 eukaryotic and 186 bacterial species to filter possible false positives in the predicted interactions. The reason for including bacterial species for the analysis was that many interactions were derived from predicted operons of bacteria. Furthermore, some orthologous proteins in experimentally identified interactions for eukaryote model organisms can be found in bacterial species.
We first built a model by training the SVM on phylogenetic profiles of positive and negative protein-protein interaction datasets. The positive data for protein-protein interactions were taken on the basis of evolutionary conservation of interactions and the number of experimental observations of interactions using the PubMed IDs. Conserved interactions or interologs that are present in more than one species are likely to be true interactions or interologs. The interactions observed by multiple experiments are also likely to be true interactions. There were 1637 interologs present in multiple species and 6043 interologs in mammals (Mus musculus, Rattus norvegicus, and Homo sapiens) with multiple PubMed IDs. After merging both datasets and removing the redundancy, the final dataset contained 7308 protein pairs corresponding to the 6348 gene pairs. The negative dataset for predicting functional linkages was assumed to be those proteins that are not co-localized in the same sub-cellular compartment . The protein localization data for Mus musculus was obtained from eSLDB . The negative dataset of protein interactions was prepared by the pairwise combination of proteins from nucleus/mitochondria and extracellular space. The negative dataset contained 708060 protein pairs corresponding to 657924 gene pairs.
The conservation score for a gene is defined as a total number of genomes in which gene homologs are found . The conservation score of a pair of genes was in turn defined as the least conservation score of any of two genes in pairs. To understand the effect of conservation score of a gene pair on prediction accuracy in the phylogenetic profile method, we used a half of randomly picked data from the positive and negative data sets for training SVM and the remaining half for testing the prediction accuracy. The process was repeated 100 times with each time by incrementing the cut-off conservation score by 1 in the negative data set while retaining the same positive data set. Figure 3 shows that the prediction accuracy of protein-protein interactions reached the maximum at the conservation score of 59.
We observed a similar trend when we plotted a true positive rate (i.e., "sensitivity") versus a false positive rate (i.e., "1 – specificity") in a receiver operator characteristic (ROC) graph [See Additional file 4]. As expected, we found that the accuracy value became maximum, when the "sensitivity" (0.82 on y-axis) and "1 – specificity" (0.13 on x-axis) values were at the point near the upper left corner. The accuracy remained constant beyond the conservation score of 59, and therefore, we considered phylogenetic profiles of genes in the negative dataset with the conservation score greater than 59. The number of negative dataset with a conservation score greater than 59 was 4454.
The best model for the prediction of protein-protein interactions was selected using the standard five-fold cross validation. Positive and negative datasets were randomly divided into five groups. One-fifth was used as a test set and remaining four-fifth was used as a training set. This was repeated five times with a different set of one-fifth used for testing each time. Additional file 5 shows the specificity, sensitivity, and accuracy at each trial of cross validation. Prediction accuracy at each trial of the cross validation was consistent with each other, indicating the homogeneity of the training dataset. The average accuracy was 84.1%. A model generated with the highest accuracy was retained as the best model, which was used to predict the positives and false positives in the interactions predicted above.
Among 7870 predicted interactions by ortholog co-occurrence in operons, 2309 were predicted as true interactions on the basis of phylogenetic profiles [See Additional file 6]. Similarly, among 58352 interactions, 41109 were predicted as true interactions using phylogenetic profiles [See Additional file 7]. In Table 1, Column 2 shows the final number of interologs in mouse from each reference organism, whereas Column 3 shows the fraction of interologs that were filtered out from each reference organism by phylogenetic profiles. The fraction of interologs that were predicted as false positives using phylogenetic profiles gradually increased with evolutionary distance between mouse and a reference organism. Exceptions were Arabidopsis thaliana and Rattus norvegicus: this might be caused by the fewer number of available interactions in the datasets.
2.4 Evaluation of predicted interactions
Evaluation of predicted interactions by frequency of co-expression and functional similarity of GO terms
Frequency of co-expression (Mean/Stddev)
Similarity of GO term (BP) Mean/Stddev)
Similarity of GO term(CC)(Mean/Stddev)
Protein – protein interaction datasets
interolog + phylogenetic profile
Interologs predicted as false positives by phylogenetic profiles
Ortholog co-occurrence in operons
Ortholog co-occurrence in operons + phylogenetic profile
2.5 Web interface for data browsing
To provide a user-friendly access to the database, we developed a WWW interface that allows one to search for the potential protein interactions for a gene or a list of gene http://lgsun.grc.nia.nih.gov/mppi/. Users can select a type of protein interaction dataset and enter names of genes as an Entrez gene ID, gene symbol, GenBank accession number for nucleotide and proteins, or NIA Mouse Gene Index U cluster IDs. Genes have also been directly linked to the NIA Mouse Gene Index . The web interface returns results as a network diagram  and a table that lists information on individual interactions, such as a method of identification, protein domains, species conservation, co-occurrence of gene symbols in PubMed abstracts, and protein localization. All the data are also available for download at our website http://lgsun.grc.nia.nih.gov/mppi/.
Interactions between proteins in Mus musculus were inferred on the basis of their orthologous interaction information in other organisms and the functional linkage information in predicted operons of bacteria. Possible false-positives in these interactions were filtered out using phylogenetic profiles on the basis of the notion that the evolutionarily conserved interactions should show similar pattern of occurrence along the genomes. Information about protein-protein interactions with high confidence will be useful to understand various processes in mammalian model organism, Mus musculus. Predicted interactions based on bacterial operons will provide useful insights into the function of mammalian mitochondrial proteins and their functional interactions. A web interface provides access to the database for a variety of investigations, including DNA microarrays and proteomics researches.
The proteomes and completely sequenced genome of bacteria and eukaryotes were downloaded from NCBI ftp site ftp://ftp.ncbi.nih.gov/genomes/. The homologous sequences of all the known open reading frames (ORFs) of Mus musculus were searched using BLASTp  against the proteome of other species with 10-4 as the cut off value. Orthologs of the Mus musculus genes were identified as the best hit by bi-directional BLASTp  searches against all proteins with 10-4 as the cut off value. It is known that, if multiple proteomes for each species are included, phylogenetic profile produces less accurate results [39, 40]. Therefore, when more than two proteome sequences for the same species were available, we selected the one that shared the maximum number of orthologs with Mus musculus. Finally there were 186 genomes of bacterial species and 26 genomes of eukaryotes species [See Additional file 2].
The SVM was trained on datasets for both positive and negative interactions. Pearson correlation co-efficient between the phylogenetic profiles of gene pairs was used as inputs to the SVM classifier. To validate the datasets for model selection and prediction accuracy, we have used five fold cross validation, in which the positives and negative datasets were randomly divided into five equal size sets. Training and testing carried out using the "svm-train" and "svm-predict" tools of the LibSvm software . In each step of cross validation four sets are used for training and remaining one for testing. In each step of testing, sensitivity, specificity and "balanced accuracy " were calculated in the following manner:
Sensitivity = (True Positives)/[(True Positives) + (False Negatives)]
Specificity = (True Negatives)/[(True Negatives) + (False Positives)]
Balanced accuracy = (Specificity + Sensitivity)/2
We used "balanced accuracy" instead of the standard overall "accuracy," because it has been reported that the accuracy becomes particularly problematic as a measure of validity, when the difference between sensitivity and specificity increases . We indeed observed this problem, when we applied the standard overall "accuracy" to the data shown in Figure 3. When the sensitivity was low and the specificity was high (1 < = gene conversion score < = 26), the overall accuracy was unreasonably high (99%) (Figure 3). In contrast, the "balanced accuracy" provided more reasonable estimates even in these cases (Figure 3; See Additional file 5).
Radial basis function (RBF) was used as a kernel of the Support vector Machine (SVM). To choose kernel parameters of SVM, we carried out "grid-search" using "grid.py" of LibSVM. In "grid-search", pairs of cost (c) and gamma (γ) were tested in each step of cross validation and one with the best cross validation accuracy was picked.
The co-expression frequency of gene pairs was calculated using a method similar to the one described previously. Mus musculus microarray datasets were downloaded from the NCBI GEO database ftp://ftp.ncbi.nih.gov/pub/geo/. The datasets with a sample number less than 11 were excluded from the analysis. Finally, there were 286 datasets. Between each possible gene pair in each dataset, Pearson correlation coefficient and its p-value was calculated using the "Pearson" function described in the Numerical recipes in C . A functional link between a gene pair was inferred if the Bonferroni corrected p value is less than 0.05.
gene expression omnibus
National Center for Biotechnology Information
National Institute on Aging
open reading frames
Radial basis function
receiver operator characteristic
support vector machine
We are grateful to the high-performance computational capabilities of the Biowulf PC/Linux cluster at the NIH http://biowulf.nih.gov for providing computational facilities. We thank Alexei Sharov for his suggestions on the Manuscript. This work was supported by the Intramural Research Program of National Institute on Aging, NIH (Z01 AG000656).
- Ito T, Tashiro K, Muta S, Ozawa R, Chiba T, Nishizawa M, Yamamoto K, Kuhara S, Sakaki Y: Toward a protein-protein interaction map of the budding yeast: A comprehensive system to examine two-hybrid interactions in all possible combinations between the yeast proteins. Proceedings of the National Academy of Sciences of the United States of America. 2000, 97: 1143-1147. 10.1073/pnas.97.3.1143.PubMedPubMed CentralView ArticleGoogle Scholar
- Gavin A-C, Bosche M, Krause R, Grandi P, Marzioch M, Bauer A, Schultz J, Rick JM, Michon A-M, Cruciat C-M: Functional organization of the yeast proteome by systematic analysis of protein complexes. Nature. 2002, 415: 141-147. 10.1038/415141a.PubMedView ArticleGoogle Scholar
- Giot L, Bader JS, Brouwer C, Chaudhuri A, Kuang B, Li Y, Hao YL, Ooi CE, Godwin B, Vitols E: A protein interaction map of Drosophila melanogaster. Science (New York, N Y). 2003, 302: 1727-1736.View ArticleGoogle Scholar
- Rual J-F, Venkatesan K, Hao T, Hirozane-Kishikawa T, Dricot A, Li N, Berriz GF, Gibbons FD, Dreze M, Ayivi-Guedehoussou N: Towards a proteome-scale map of the human protein-protein interaction network. Nature. 2005, 437: 1173-1178. 10.1038/nature04209.PubMedView ArticleGoogle Scholar
- Stelzl U, Worm U, Lalowski M, Haenig C, Brembeck FH, Goehler H, Stroedicke M, Zenkner M, Schoenherr A, Koeppen S: A human protein-protein interaction network: a resource for annotating the proteome. Cell. 2005, 122: 957-968. 10.1016/j.cell.2005.08.029.PubMedView ArticleGoogle Scholar
- Arifuzzaman M, Maeda M, Itoh A, Nishikata K, Takita C, Saito R, Ara T, Nakahigashi K, Huang H-C, Hirai A: Large-scale identification of protein-protein interaction of Escherichia coli K-12. Genome research. 2006, 16: 686-691. 10.1101/gr.4527806.PubMedPubMed CentralView ArticleGoogle Scholar
- Stark C, Breitkreutz B-J, Reguly T, Boucher L, Breitkreutz A, Tyers M: BioGRID: a general repository for interaction datasets. Nucleic acids research. 2006, 34: D535-539. 10.1093/nar/gkj109.PubMedPubMed CentralView ArticleGoogle Scholar
- Kerrien S, Alam-Faruque Y, Aranda B, Bancarz I, Bridge A, Derow C, Dimmer E, Feuermann M, Friedrichsen A, Huntley R: IntAct–open source resource for molecular interaction data. Nucleic acids research. 2007, 35: D561-565. 10.1093/nar/gkl958.PubMedPubMed CentralView ArticleGoogle Scholar
- Chatr-aryamontri A, Ceol A, Palazzi LM, Nardelli G, Schneider MV, Castagnoli L, Cesareni G: MINT: the Molecular INTeraction database. Nucleic acids research. 2007, 35: D572-574. 10.1093/nar/gkl950.PubMedPubMed CentralView ArticleGoogle Scholar
- Salwinski L, Eisenberg D: In silico simulation of biological network dynamics. Nature biotechnology. 2004, 22: 1017-1019. 10.1038/nbt991.PubMedView ArticleGoogle Scholar
- Vastrik I, D'Eustachio P, Schmidt E, Joshi-Tope G, Gopinath G, Croft D, de Bono B, Gillespie M, Jassal B, Lewis S: Reactome: a knowledge base of biologic pathways and processes. Genome Biol. 2007, 8: R39-10.1186/gb-2007-8-3-r39.PubMedPubMed CentralView ArticleGoogle Scholar
- Matthews LR, Vaglio P, Reboul J, Ge H, Davis BP, Garrels J, Vincent S, Vidal M: Identification of potential interaction networks using sequence-based searches for conserved protein-protein interactions or "interologs". Genome research. 2001, 11: 2120-2126. 10.1101/gr.205301.PubMedPubMed CentralView ArticleGoogle Scholar
- Yu H, Luscombe NM, Lu HX, Zhu X, Xia Y, Han J-DJ, Bertin N, Chung S, Vidal M, Gerstein M: Annotation transfer between genomes: protein-protein interologs and protein-DNA regulogs. Genome research. 2004, 14: 1107-1118. 10.1101/gr.1774904.PubMedPubMed CentralView ArticleGoogle Scholar
- Dandekar T, Snel B, Huynen M, Bork P: Conservation of gene order: a fingerprint of proteins that physically interact. Trends in biochemical sciences. 1998, 23: 324-328. 10.1016/S0968-0004(98)01274-2.PubMedView ArticleGoogle Scholar
- Overbeek R, Fonstein M, D'Souza M, Pusch GD, Maltsev N: The use of gene clusters to infer functional coupling. Proceedings of the National Academy of Sciences of the United States of America. 1999, 96: 2896-2901. 10.1073/pnas.96.6.2896.PubMedPubMed CentralView ArticleGoogle Scholar
- Yellaboina S, Goyal K, Mande SC: Inferring genome-wide functional linkages in E. coli by combining improved genome context methods: comparison with high-throughput experimental data. Genome research. 2007, 17: 527-535. 10.1101/gr.5900607.PubMedPubMed CentralView ArticleGoogle Scholar
- Janga SC, Collado-Vides J, Moreno-Hagelsieb G: Nebulon: a system for the inference of functional relationships of gene products from the rearrangement of predicted operons. Nucleic Acids Res. 2005, 33: 2521-2530. 10.1093/nar/gki545.PubMedPubMed CentralView ArticleGoogle Scholar
- Enright AJ, Iliopoulos I, Kyrpides NC, Ouzounis CA: Protein interaction maps for complete genomes based on gene fusion events. Nature. 1999, 402: 86-90. 10.1038/47056.PubMedView ArticleGoogle Scholar
- Pellegrini M, Marcotte EM, Thompson MJ, Eisenberg D, Yeates TO: Assigning protein functions by comparative genome analysis: protein phylogenetic profiles. Proceedings of the National Academy of Sciences of the United States of America. 1999, 96: 4285-4288. 10.1073/pnas.96.8.4285.PubMedPubMed CentralView ArticleGoogle Scholar
- Kim Y, Koyuturk M, Topkara U, Grama A, Subramaniam S: Inferring functional information from domain co-evolution. Bioinformatics. 2006, 22: 40-49. 10.1093/bioinformatics/bti723.PubMedView ArticleGoogle Scholar
- Pagel P, Wong P, Frishman D: A domain interaction map based on phylogenetic profiling. J Mol Biol. 2004, 344: 1331-1346. 10.1016/j.jmb.2004.10.019.PubMedView ArticleGoogle Scholar
- Pazos F, Valencia A: In silico two-hybrid system for the selection of physically interacting protein pairs. Proteins. 2002, 47: 219-227. 10.1002/prot.10074.PubMedView ArticleGoogle Scholar
- von Mering C, Krause R, Snel B, Cornell M, Oliver SG, Fields S, Bork P: Comparative assessment of large-scale data sets of protein-protein interactions. Nature. 2002, 417: 399-403. 10.1038/nature750.PubMedView ArticleGoogle Scholar
- Mika S, Rost B: Protein-protein interactions more conserved within species than across species. PLoS computational biology. 2006, 2: e79-10.1371/journal.pcbi.0020079.PubMedPubMed CentralView ArticleGoogle Scholar
- Huang TW, Lin CY, Kao CY: Reconstruction of human protein interolog network using evolutionary conserved network. BMC Bioinformatics. 2007, 8: 152-10.1186/1471-2105-8-152.PubMedPubMed CentralView ArticleGoogle Scholar
- Persico M, Ceol A, Gavrila C, Hoffmann R, Florio A, Cesareni G: HomoMINT: an inferred human network based on orthology mapping of protein interactions discovered in model organisms. BMC Bioinformatics. 2005, 6 (Suppl 4): S21-10.1186/1471-2105-6-S4-S21.PubMedPubMed CentralView ArticleGoogle Scholar
- Patil A, Nakamura H: Filtering high-throughput protein-protein interaction data using a combination of genomic features. BMC Bioinformatics. 2005, 6: 100-10.1186/1471-2105-6-100.PubMedPubMed CentralView ArticleGoogle Scholar
- Mahdavi MA, Lin YH: False positive reduction in protein-protein interaction predictions using gene ontology annotations. BMC Bioinformatics. 2007, 8: 262-10.1186/1471-2105-8-262.PubMedPubMed CentralView ArticleGoogle Scholar
- Mishra GR, Suresh M, Kumaran K, Kannabiran N, Suresh S, Bala P, Shivakumar K, Anuradha N, Reddy R, Raghavan TM: Human protein reference database–2006 update. Nucleic acids research. 2006, 34: D411-414. 10.1093/nar/gkj141.PubMedPubMed CentralView ArticleGoogle Scholar
- Altschul SF, Madden TL, Schaffer AA, Zhang J, Zhang Z, Miller W, Lipman DJ: Gapped BLAST and PSI-BLAST: a new generation of protein database search programs. Nucleic Acids Res. 1997, 25: 3389-3402. 10.1093/nar/25.17.3389.PubMedPubMed CentralView ArticleGoogle Scholar
- Howe CJ: Cellular evolution: what's in a mitochondrion?. Curr Biol. 2008, 18: R429-431. 10.1016/j.cub.2008.04.007.PubMedView ArticleGoogle Scholar
- Jansen R, Yu H, Greenbaum D, Kluger Y, Krogan NJ, Chung S, Emili A, Snyder M, Greenblatt JF, Gerstein M: A Bayesian networks approach for predicting protein-protein interactions from genomic data. Science. 2003, 302: 449-453. 10.1126/science.1087361.PubMedView ArticleGoogle Scholar
- Pierleoni A, Martelli PL, Fariselli P, Casadio R: eSLDB: eukaryotic subcellular localization database. Nucleic Acids Res. 2007, 35: D208-212. 10.1093/nar/gkl775.PubMedPubMed CentralView ArticleGoogle Scholar
- Enault F, Suhre K, Abergel C, Poirot O, Claverie JM: Annotation of bacterial genomes using improved phylogenomic profiles. Bioinformatics (Oxford, England). 2003, 19 (Suppl 1): i105-107. 10.1093/bioinformatics/btg1013.View ArticleGoogle Scholar
- Sun J, Xu J, Liu Z, Liu Q, Zhao A, Shi T, Li Y: Refined phylogenetic profiles method for predicting protein-protein interactions. Bioinformatics (Oxford, England). 2005, 21: 3409-3415. 10.1093/bioinformatics/bti532.View ArticleGoogle Scholar
- Frohlich H, Speer N, Poustka A, Beissbarth T: GOSim–an R-package for computation of information theoretic GO similarities between terms and gene products. BMC Bioinformatics. 2007, 8: 166-10.1186/1471-2105-8-166.PubMedPubMed CentralView ArticleGoogle Scholar
- Sharov AA, Dudekula DB, Ko MS: Genome-wide assembly and analysis of alternative transcripts in mouse. Genome Res. 2005, 15: 748-754. 10.1101/gr.3269805.PubMedPubMed CentralView ArticleGoogle Scholar
- Mrowka R: A Java applet for visualizing protein-protein interaction. Bioinformatics. 2001, 17: 669-671. 10.1093/bioinformatics/17.7.669.PubMedView ArticleGoogle Scholar
- Karimpour-Fard A, Hunter L, Gill RT: Investigation of factors affecting prediction of protein-protein interaction networks by phylogenetic profiling. BMC Genomics. 2007, 8: 393-10.1186/1471-2164-8-393.PubMedPubMed CentralView ArticleGoogle Scholar
- Moreno-Hagelsieb G, Janga SC: Operons and the effect of genome redundancy in deciphering functional relationships using phylogenetic profiles. Proteins. 2008, 70: 344-352. 10.1002/prot.21564.PubMedView ArticleGoogle Scholar
- Chang C-C, Lin C-J: LIBSVM: a library for support vector machines. 2001Google Scholar
- Velez DR, White BC, Motsinger AA, Bush WS, Ritchie MD, Williams SM, Moore JH: A balanced accuracy function for epistasis modeling in imbalanced datasets using multifactor dimensionality reduction. Genet Epidemiol. 2007, 31: 306-315. 10.1002/gepi.20211.PubMedView ArticleGoogle Scholar
- Alberg AJ, Park JW, Hager BW, Brock MV, Diener-West M: The use of "overall accuracy" to evaluate the validity of screening or diagnostic tests. J Gen Intern Med. 2004, 19: 460-465. 10.1111/j.1525-1497.2004.30091.x.PubMedPubMed CentralView ArticleGoogle Scholar
- Stuart JM, Segal E, Koller D, Kim SK: A gene-coexpression network for global discovery of conserved genetic modules. Science. 2003, 302: 249-255. 10.1126/science.1087447.PubMedView ArticleGoogle Scholar
- Press WH, Teukolsky SA, Vetterling WT, Flannery BP: Numerical Recipes in C: The Art of Scientific Computing. 1993, Cambridge University PressGoogle Scholar
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.