- Methodology article
- Open Access
Improved homology-driven computational validation of protein-protein interactions motivated by the evolutionary gene duplication and divergence hypothesis
© Frech et al; licensee BioMed Central Ltd. 2009
- Received: 09 June 2008
- Accepted: 19 January 2009
- Published: 19 January 2009
Protein-protein interaction (PPI) data sets generated by high-throughput experiments are contaminated by large numbers of erroneous PPIs. Therefore, computational methods for PPI validation are necessary to improve the quality of such data sets. Against the background of the theory that most extant PPIs arose as a consequence of gene duplication, the sensitive search for homologous PPIs, i.e. for PPIs descending from a common ancestral PPI, should be a successful strategy for PPI validation.
To validate an experimentally observed PPI, we combine FASTA and PSI-BLAST to perform a sensitive sequence-based search for pairs of interacting homologous proteins within a large, integrated PPI database. A novel scoring scheme that incorporates both quality and quantity of all observed matches allows us (1) to consider also tentative paralogs and orthologs in this analysis and (2) to combine search results from more than one homology detection method. ROC curves illustrate the high efficacy of this approach and its improvement over other homology-based validation methods.
New PPIs are primarily derived from preexisting PPIs and not invented de novo. Thus, the hallmark of true PPIs is the existence of homologous PPIs. The sensitive search for homologous PPIs within a large body of known PPIs is an efficient strategy to separate biologically relevant PPIs from the many spurious PPIs reported by high-throughput experiments.
- Receiver Operating Characteristic Curve
- Homology Detection
- Weak Homolog
- Gold Standard Positive
- Gold Standard Negative
Physical interactions between proteins, commonly referred to as protein-protein interactions (PPIs), occur at every level of cell function to elaborate the organism's phenotype. The study of PPIs is therefore of great interest and is helping to reveal basic molecular mechanisms of many diseases. High-throughput screening methods have given insight into hundreds of thousands of potential PPIs in several organisms. However, a major disadvantage of high-throughput approaches is their high rate of false-positive PPIs, i.e. erroneously reported PPIs that do not occur in vivo [1–7].
The development and implementation of computational methods for the validation of experimentally determined PPIs is therefore an important goal in bioinformatics today. Common approaches include determining intersections between different high-throughput PPI data sets , incorporating protein annotation data [5, 8], analyzing expression profiles [4, 9–12], investigating topological criteria of PPI networks [13–17], and inspecting patterns of co-evolution .
Firstly, we draw the reader's attention to the duplication-divergence hypothesis of PPI evolution, i.e. the idea that extant PPIs primarily originate from gene duplications, the homologs diverging over time. If PPIs share common evolutionary ancestry, which is what this hypothesis suggests, then this ancestry reaches far into the evolutionary past. Consequently, homology-based PPI validation should investigate also diverged homologs and not only similar proteins.
Secondly, motivated by this idea, we propose an improved, sequence-based procedure for homology-based PPI validation. Unlike previously published, mostly binary validation schemes that deem a questioned PPI as biologically relevant as soon as a single homologous PPI is found, we follow a similar approach as Jonsson et al.  and compute a confidence score that takes into account both the quality and quantity of all identified homologous PPIs. The assignment of higher scores to high-quality hits and of lower scores to low-quality hits allows us to extend the search for homologous PPIs from reliable homologs to highly putative paralogs and orthologs with E-values up to 10. In addition to similar scoring schemes proposed before, we normalize and combine scores obtained from different homology search strategies.
Thirdly, we demonstrate the high efficacy of homology-based validation when carried out on large PPI data sets. A comprehensive data set of known physical binary PPIs from six PPI source databases is compiled, comprising 135,276 PPIs from 20 different organisms. This is, to the best of our knowledge, the largest collection of PPIs that has been used so far in this kind of analysis. Based on Receiver Operating Characteristic (ROC) curves it is shown that the new approach improves over previous methods for homology-based PPI validation.
Duplication-Divergence Hypothesis of PPI Evolution
Gene duplication is a ubiquitous mechanism in molecular evolution and the principal source of biological innovation, producing new proteins and novel functional domains [26–30]. Here, we follow the idea that the duplication of genetic material coupled with subsequent divergence is also the dominant mechanism for the development of novel PPIs . This hypothesis is supported by both theoretical models [32–34] and empirical evidence [35–40]. A brief review of papers supporting the duplication-divergence hypothesis can be found in Additional file 1.
Implications of the Duplication-Divergence Model for Homology-Based PPI Validation
The duplication-divergence model of PPI evolution as shown in Figure 2 suggests that most biologically relevant PPIs descend from a common ancestral PPI, i.e. the PPIs are homologous to each other. This allows assessing the plausibility of an experimentally determined PPI as follows: for a true-positive PPI, one expects to see many homologous PPIs, whereas for a false-positive PPI this should be less likely. A sensitive search for homologous PPIs should thus, in principle, be able to filter out large numbers of false-positive PPIs from experimental PPI data sets while retaining the bulk of true-positive PPIs. However, both the incompleteness of today's PPI data sets and the fact that common ancestry is often elusive represent major practical obstacles along the way.
For assessing the validity of an experimentally predetermined PPI, also 'weak' homologous PPIs can be informative ('weak' in the sense of 'weak signal of homology'). Their incidence might support or contradict the idea that a questioned PPI evolved by duplication-divergence, which in turn can strengthen or weaken the position that a PPI is biologically relevant. However, the value of weak homologous PPIs for PPI prediction is limited: one cannot infer from a given pair of interacting proteins that very distant homologs interact as well. In the majority of cases this prediction would be simply wrong, because due to divergence most duplicated PPIs are eventually lost. PPIs inferred by homology are thus only trustworthy if protein similarity is high [41, 42] or if these PPIs are supported by complementary data .
Weak Homologous Interactions – Signal or Noise?
Not surprisingly, the first characteristic, the existence of at least one homologous PPI, is a highly reliable signal for GSP PPIs when sequence similarity is high. For example, almost every fifth PPI taken out of the GSP data set (18%) has a homologous PPI with an E-value lower than 10-100 (Figure 3A). By contrast, the existence of such high-quality homologs is extremely unlikely for a GSN PPI (0.25%). The signal remains intact with very low levels of sequence similarity: within the last E-value window (ranging from 3 to 10) the probability of observing a homologous PPI for a GSP PPI remains still twice as high (63%) as for a GSN PPI (30%).
The distribution of homologous PPIs reveals the second interesting characteristic of GSP PPIs: they tend to accumulate large numbers of homologous PPIs. According to Figure 3A, in all windows with E-values greater than 10-20, about 25% of GSP PPIs have more than 10 homologous PPIs; for GSN PPIs, this percentage never exceeds 8%. Thus, for many PPIs the existence of a large number of homologous PPIs is more conclusive than the existence of at least one homologous PPI, especially when sequence similarity is low: whereas about twice as many GSP PPIs than GSN PPIs have at least one homologous PPI within the last E-value window, almost four times as many GSP PPIs (21.4%) than GSN PPIs (5.8%) have between 10 and 100 homologs, and more than five times as many GSP PPIs (6.8%) than GSN PPIs (1.3%) have more than 100 homologs.
Both characteristics are observed independently of the fact whether only paralogous (Figure 3B) or only orthologous PPIs (Figure 3C) are investigated. For lower numbers of homologous PPIs, stronger signals on the paralogous data set are obtained than on the orthologous data set, which is consistent with the finding that PPIs seem to be more conserved within species than across species . Interestingly, very large numbers (>100) of homologous PPIs are observed within the orthologous data set, most likely due to an increased number of gene duplications in higher eukaryotes.
We conclude that weak homologous PPIs are indeed an observable and distinguishing characteristic of biologically relevant PPIs, especially if they are observed in increased numbers. Consequently, weak homologous PPIs should be considered by homology-based PPI validation schemes.
The Small Scale gold standard performs worst, which might reflect differences in the quality of the data sets. The Multiple Evidence gold standard can be considered of highest quality, because detection of a PPI with different experimental methods is a very reliable indicator of its existence . Indeed, this data set achieves the best performance up to a TPR of 70%. The MIPS gold standard contains PPIs audited by human experts and is thus very trustworthy as well, although a slightly poorer performance is seen on this data set. PPIs of the Small Scale gold standard are not reviewed manually and thus its reliability might to a large degree reflect the quality of the automated text mining tools that are frequently used to extract them from the scientific literature. Because these tools are error-prone , the Small Scale gold standard might contain more spurious PPIs than the other two data sets.
Note that although the class distribution in the gold standard data sets is skewed, i.e. the Random GSN data set is about 50 times larger than each GSP data set, this does not affect the overall ROC curve . In fact, we observed the same overall ROC curve on a balanced data set where the number of randomly chosen GSNs roughly equals the number of GSPs (data not shown). ROC curves as shown in Figure 4 are ideal to illustrate the overall performance of a classifier, but do not make suggestions about which specific score threshold should be applied to classify a PPI as true or false. This decision depends on the TPR and FPR one is willing to accept. Supplementary Figure 1 shows selected score thresholds and their associated TPRs and FPRs.
Comparison of Homology-Based Validation Schemes
The FASTA-based binary validation scheme shows remarkably high specificity, even at high E-values. For example, inclusion of homologous PPIs with E-values between 10-4 and 10 results in an increase of the TPR of almost 25% (from 48.5% to 72.1%), whereas the FPR grows only by 13.5% within the same interval, remaining below 16%. Considering the low sequence similarity and the probable existence of many spurious hits at E-values up to 10, this find is remarkable. By contrast, the PSI-BLAST-based binary validation scheme is less specific (even at low E-values), but much more sensitive: almost 87% of the GSP PPIs have at least one homologous PPI identified by PSI-BLAST (E-value ≤ 10). In addition to FASTA and PSI-BLAST, also BLAST was evaluated for homology detection (data not shown). In comparison to FASTA, no noticeable difference in performance was observed except for a slight decrease in maximum sensitivity (about 2% lower than with FASTA). Our scoring scheme, represented by the blue curve (squares), combines evidence from homologous PPIs found by FASTA and PSI-BLAST and clearly outperforms both individual binary validation schemes. For example, at a TPR of 70% it produces 4% fewer false-positives than the FASTA-based binary approach, and about 6% fewer false-positives than the PSI-BLAST-based binary validation scheme.
Previous homology-based validation schemes suggested different parameter settings. Saeed and Deane  used BLAST with an E-value up to 10-4 to identify homologous PPIs and evaluated a TPR of 63% at an FPR of 7%. If this setting is transferred to FASTA and applied on our data sets, a TPR of 48.5% at an FPR of 2.3% is observed. Our scoring scheme, by contrast, produces only 1.5% false-positives at the same level of sensitivity (48.5%). The difference in FPR increases for higher levels of sensitivity, illustrating the additional value from incorporation of multiple methods for homology detection and from consideration of weak homologs. Patil and Nakamura  used PSI-BLAST with an E-value up to 10-8 and reported a TPR of 89.7% at an FPR of 37.1% for their gold standards. On our data set the same parameter setting results in a significantly worse TPR of 70.1% at an FPR of 16.1%. Again, the scoring scheme outperforms and achieves a reduced FPR of 10% at the same level of sensitivity (70.1%). It is noteworthy that our exclusively sequence-based scoring scheme produces a superior ROC curve than Patil and Nakamura's Bayesian network approach, which incorporates three genomic features instead of one (sequence, structure and annotation information). This underscores again the potential efficacy of homology-based methods. No published PSI-BLAST parameters were found in the paper from Deane et al. , and thus the performance of this method was not assessed.
Contribution of Weak Homologs
The inclusion of weak homologous PPIs contributes positively to the overall classification performance. For example, the restriction of the analysis to homologous PPIs with an E-value below 10-10 results in a maximum TPR of 69% at an FPR of 14.5%. When homologs with E-values up to 1 are considered, the same sensitivity is achieved at a significantly reduced FPR of 10%. Another increase of the E-value threshold up to 10 leads to a further reduction of the FPR by 1%.
Note that a similar effect cannot be observed for the classic, binary validation schemes, where less stringent E-value thresholds increase sensitivity but decrease specificity (Figure 5). This emphasizes the value of the scoring approach: it finds evidence for biologically relevant PPIs among weaker homologs without the compromise of an increased rate of false-positives.
Although the additional benefit resulting from the inclusion of weak homologs with an E-value above 1 is rather low on this data set, we expect it to increase for data sets where high-quality homologs are not at hand. Yeast is comparably well investigated, with approximately 50% of its estimated 40,000 to 75,000 PPIs known . As a consequence, most of its biologically relevant PPIs have homologous PPIs among high-quality paralogs, and matches among weak homologs add little extra value to the overall score. This situation is different from most other organisms where interactome coverage is far below 50% and where weak paralogs and orthologs are often the only possibility to validate a questioned PPI.
Knowledge of PPIs is key to understanding cell function. Although experimental high-throughput PPI detection techniques are now making it possible to catalogue all PPIs of a cell, the notoriously high error rates of these methods are a major obstacle to achieving this ambitious goal. Computational methods that can efficiently separate the PPI wheat from the chaff are therefore highly desirable.
We think that recent insights into the evolution of PPIs, in particular the duplication-divergence hypothesis, might be crucial to this endeavor. Nature is a tinkerer, not an inventor . For PPIs this means that new PPIs are primarily derived from preexisting PPIs rather than invented de novo. Consequently, most true-positive PPIs must have homologous PPIs, not only among highly similar proteins, but also among distantly related proteins. This important characteristic of biologically relevant PPIs should, in principle, allow successful discrimination between true and false PPIs.
In light of this consideration, homology-based validation techniques seem promising, but have not gained much attention so far. Literature searches revealed only five papers that proposed a homology-based technique to validate experimental PPIs on a large scale, only three of which presented a critical performance assessment. Presumably this reflects the fact that homology-based validation requires having at hand a set of PPIs among homologous proteins, when few such PPIs have been known. However, with more and more PPIs now being reported from high-throughput experiments, this limitation is no longer factor.
In this paper, we assembled a large PPI data set to reassess the performance of homology-based PPI validation. It was shown that the classic, binary validation technique is efficient on such data sets, but can be further improved by using multiple methods for homology detection and more remote homologs to complement close homologs.
We expect the findings to be most relevant in situations where interactions among assured paralogs or orthologs are not at hand and thus traditional homology-based validation is not an option. Existing PPI databases could use the proposed method to reduce their number of false-positives without losing too many true-positives, especially within well explored model organisms. Other prospective applications include the elucidation of physically interacting proteins from known protein complexes, or the validation of in silico predicted PPIs in cases where homology was not used as a criterion for prediction in advance. Prospective improvements may involve more sophisticated methods for homology detection (e.g. Profile-HMMs), identification of PPI-mediating protein features (e.g. interacting domains) prior to homology detection to refine the selection of homologs, and an assessment of the statistical significance (P-values) of computed scores to obtain an intuitive measure of a PPI's validity.
Database Search for Homologous PPIs
The modular architecture of proteins implies that a protein has not just one distinct evolutionary trajectory, but one for each biological feature it contains . Since in general PPI-mediating features (e.g. the domains) are unknown, one cannot selectively examine only trajectories of relevance. One possibility is to examine evolutionary trajectories of all protein features, but this comes with an increased risk of detecting PPIs that are not truly homologous. Also, if one wishes to include weak paralogs and orthologs in the analysis to capture gene duplication events that happened long time ago, methods for homology detection become inaccurate and produce spurious hits – another source of false homologous PPIs. To maximize both sensitivity and specificity despite these difficulties, we opt for a large-scale, sequence-based screening procedure in combination with a scoring scheme. Given an experimentally determined PPI, both FASTA  and PSI-BLAST  are used to search for homologous PPIs. FASTA supplies more reliable results for closely related proteins, while PSI-BLAST is more sensitive for remote relationships . To further increase the sensitivity of the method, local sequence similarities with E-values up to 10 are considered. This produces many spurious hits, and thus traditional homology-based PPI validation techniques that simply check for the existence of a single homologous PPI become misleading (compare Figure 3). We therefore follow a similar approach as Jonsson et al.  and apply a scoring scheme that weighs each match according to its sequence similarity: low E-values score high, and high E-values score low. Thus high scores can result from few high-quality hits but also from numerous low-quality hits.
Homologous PPIs are searched within a subset of the Protein Interaction and Molecule Search (PRIMOS) database http://primos.fh-hagenberg.at, release BETA-2.7/2007–04 . This subset consists of 135,276 redundancy-removed, physical binary PPIs between 42,288 proteins from 20 organisms, imported from six primary PPI databases [52–57] (see Additional file 1). We used both FASTA (fasta34.exe, v3.4) and PSI-BLAST (blastpgp.exe, v2.2.16) to determine homologs for all proteins of our gold standard data sets. The search space for homologous proteins was restricted to the set of 42,288 proteins with known PPIs. According to Figure 1, we considered a homologous PPI as an interaction found between a pair of homologous proteins. E-value thresholds for both programs were set to 10, the ktup parameter of FASTA was set to 1, and the number of iterations for PSI-BLAST was set to 10. All other program parameters were left default.
Gold Standard Data Sets
Four gold standard positive (GSP) data sets and one gold standard negative (GSN) data set with PPIs from Saccharomyces cerevisiae are used for performance assessment. Yeast is relatively well-studied, which allows being rather stringent in the selection of the GSP data sets. In addition, yeast has already been used numerous times in similar studies, which eases the comparison with previous results.
The MIPS GSP data set comprises 1,541 physical binary PPIs obtained from the Comprehensive Yeast Genome Database (CYGD) . This database is considered as a high-quality resource for yeast PPIs and is frequently used as a gold standard reference set. PPIs reported by high-throughput experiments are excluded from this data set (see Additional file 1). The Multiple Evidence GSP data set consists of 393 PPIs reported by at least two experimental methods and in at least two different publications. As an additional criterion, only publications imported from one PPI database are considered. For example, if a PPI is reported from a publication contained in DIP and MINT, it will be excluded from the data set. If DIP is the only source database for this PPI, the PPI will be included. In an integrated dataset compiled from multiple source data sets this procedure reduces the risk that duplicate PPIs are regarded as homologous. The Small Scale data set consists of yeast PPIs reported by 'small-scale' experiments and contains 902 PPIs. Only PPIs of publications with up to three reported PPIs are considered. To minimize the risk of duplicate PPIs, publications imported from more than one primary PPI database are excluded (same procedure as for the Multiple Evidence GSP). The three GSP data sets overlap only to a low degree: just 8 PPIs are common to all three data sets, 25 between MIPS and Small Scale, 86 between Small Scale and Multiple Evidence, and 10 between MIPS and Multiple Evidence. The Combined GSP data set contains all PPIs from the previous three GSP data sets (2,723 PPIs in total).
The Random GSN PPI data set was generated by randomly selecting 50,000 protein pairs out of 7,058 yeast proteins (UniProt  release 10.0, downloaded on March 29, 2007) that were not found interacting within the PRIMOS database. A randomly selected data set is not completely free of real PPIs, but has no selection bias, for example towards protein pairs with different molecular functions . The amount of real PPIs within such a randomly selected GSN data set should be generally low at about 0.25% .
where O is the set of organisms with known experimental PPIs in the PRIMOS database, H a (o) and H b (o) denote the sets of proteins from organism o that are homologous to protein a and b, respectively. If there is experimental evidence for an interaction between homolog h a and homolog h b in the PRIMOS database, a score proportional to their sequence similarity is added to an overall sum.
where Smax(a, b) is defined as S(a, b) with all h a assumed as interacting with all h b . This scales the score to values ranging from 0 (minimum score) to 1 (maximum score).
This research was supported by the Austrian Research Promotion Agency (FFG) under the FHplus program and has been financed by the Austrian government (BMVIT and BMBWK) as well as by our co-financing partner Salzburger Landeskliniken GmbH. The authors thank all partners and colleagues that contributed to this project with their work, especially Wolfgang Straßer and Doris Siegl, who developed the underlying PRIMOS system. Any opinions, findings, and conclusions or recommendations in this paper are those of the authors and do not necessarily represent the views of the research sponsors.
- Mrowka R, Patzak A, Herzel H: Is there a bias in proteome research? Genome Res 2001, 11(12):1971–1973. 10.1101/gr.206701View ArticlePubMedGoogle Scholar
- Legrain P, Wojcik J, Gauthier JM: Protein-protein interaction maps: a lead towards cellular functions. Trends Genet 2001, 17(6):346–352. 10.1016/S0168-9525(01)02323-XView ArticlePubMedGoogle 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(6887):399–403. 10.1038/nature750View ArticlePubMedGoogle Scholar
- Deane CM, Salwinski L, Xenarios I, Eisenberg D: Protein interactions: two methods for assessment of the reliability of high throughput observations. Mol Cell Proteomics 2002, 1(5):349–356. 10.1074/mcp.M100037-MCP200View ArticlePubMedGoogle Scholar
- Sprinzak E, Sattath S, Margalit H: How reliable are experimental protein-protein interaction data? J Mol Biol 2003, 327(5):919–923. 10.1016/S0022-2836(03)00239-0View ArticlePubMedGoogle Scholar
- Gilchrist MA, Salter LA, Wagner A: A statistical framework for combining and interpreting proteomic datasets. Bioinformatics 2004, 20(5):689–700. 10.1093/bioinformatics/btg469View ArticlePubMedGoogle Scholar
- Edwards AM, Kus B, Jansen R, Greenbaum D, Greenblatt J, Gerstein M: Bridging structural biology and genomics: assessing protein interaction data with known complexes. Trends Genet 2002, 18(10):529–536. 10.1016/S0168-9525(02)02763-4View ArticlePubMedGoogle Scholar
- Patil A, Nakamura H: Filtering high-throughput protein-protein interaction data using a combination of genomic features. BMC Bioinformatics 2005, 6: 100. 10.1186/1471-2105-6-100PubMed CentralView ArticlePubMedGoogle Scholar
- Jansen R, Greenbaum D, Gerstein M: Relating whole-genome expression data with protein-protein interactions. Genome Res 2002, 12: 37–46. 10.1101/gr.205602PubMed CentralView ArticlePubMedGoogle Scholar
- Kemmeren P, van Berkum NL, Vilo J, Bijma T, Donders R, Brazma A, Holstege FCP: Protein interaction verification and functional annotation by integrated analysis of genome-scale data. Mol Cell 2002, 9(5):1133–1143. 10.1016/S1097-2765(02)00531-2View ArticlePubMedGoogle Scholar
- Deng M, Sun F, Chen T: Assessment of the reliability of protein-protein interactions and protein function prediction. Pac Symp Biocomput 2003, 140–151.Google Scholar
- Tirosh I, Barkai N: Computational verification of protein-protein interactions by orthologous co-expression. BMC Bioinformatics 2005, 6: 40. 10.1186/1471-2105-6-40PubMed CentralView ArticlePubMedGoogle Scholar
- Saito R, Suzuki H, Hayashizaki Y: Construction of reliable protein-protein interaction networks with a new interaction generality measure. Bioinformatics 2003, 19(6):756–763. 10.1093/bioinformatics/btg070View ArticlePubMedGoogle Scholar
- Goldberg DS, Roth FP: Assessing experimentally derived interactions in a small world. Proc Natl Acad Sci USA 2003, 100(8):4372–4376. 10.1073/pnas.0735871100PubMed CentralView ArticlePubMedGoogle Scholar
- Bader JS, Chaudhuri A, Rothberg JM, Chant J: Gaining confidence in high-throughput protein interaction networks. Nat Biotechnol 2004, 22: 78–85. 10.1038/nbt924View ArticlePubMedGoogle Scholar
- Chen J, Hsu W, Lee ML, Ng SK: Discovering reliable protein interactions from high-throughput experimental data using network topology. Artif Intell Med 2005, 35(1–2):37–47. 10.1016/j.artmed.2005.02.004View ArticlePubMedGoogle Scholar
- Pei P, Zhang A: A topological measurement for weighted protein interaction network. Proc IEEE Comput Syst Bioinform Conf 2005, 268–278.Google Scholar
- Tan SH, Zhang Z, Ng SK: Automated Detection and Validation of Interaction by Co-Evolution. Nucleic Acids Res 2004, (32 Web Server):W69-W72. 10.1093/nar/gkh471Google 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 Res 2001, 11(12):2120–2126. 10.1101/gr.205301PubMed CentralView ArticlePubMedGoogle Scholar
- Walhout AJ, Sordella R, Lu X, Hartley JL, Temple GF, Brasch MA, Thierry-Mieg N, Vidal M: Protein interaction mapping in C. elegans using proteins involved in vulval development. Science 2000, 287(5450):116–122. 10.1126/science.287.5450.116View ArticlePubMedGoogle Scholar
- Remm M, Storm CE, Sonnhammer EL: Automatic clustering of orthologs and in-paralogs from pairwise species comparisons. J Mol Biol 2001, 314(5):1041–1052. 10.1006/jmbi.2000.5197View ArticlePubMedGoogle Scholar
- Suthram S, Shlomi T, Ruppin E, Sharan R, Ideker T: A direct comparison of protein interaction confidence assignment schemes. BMC Bioinformatics 2006, 7: 360. 10.1186/1471-2105-7-360PubMed CentralView ArticlePubMedGoogle Scholar
- Saeed R, Deane C: An assessment of the uses of homologous interactions. Bioinformatics 2007.Google Scholar
- Brown KR, Jurisica I: Online predicted human interaction database. Bioinformatics 2005, 21(9):2076–2082. 10.1093/bioinformatics/bti273View ArticlePubMedGoogle Scholar
- Jonsson PF, Cavanna T, Zicha D, Bates PA: Cluster analysis of networks generated through homology: automatic identification of important protein communities involved in cancer metastasis. BMC Bioinformatics 2006, 7: 2. 10.1186/1471-2105-7-2PubMed CentralView ArticlePubMedGoogle Scholar
- Zhang J: Evolution by gene duplication: an update. Trends in Ecology and Evolution 2003, 18(6):292–298. 10.1016/S0169-5347(03)00033-8View ArticleGoogle Scholar
- Gough J, Karplus K, Hughey R, Chothia C: Assignment of homology to genome sequences using a library of hidden Markov models that represent all proteins of known structure. J Mol Biol 2001, 313(4):903–919. 10.1006/jmbi.2001.5080View ArticlePubMedGoogle Scholar
- Todd AE, Orengo CA, Thornton JM: Evolution of function in protein superfamilies, from a structural perspective. J Mol Biol 2001, 307(4):1113–1143. 10.1006/jmbi.2001.4513View ArticlePubMedGoogle Scholar
- Murzin AG: How far divergent evolution goes in proteins. Curr Opin Struct Biol 1998, 8(3):380–387. 10.1016/S0959-440X(98)80073-0View ArticlePubMedGoogle Scholar
- Ohno S: Evolution by gene duplication. Springer-Verlag; 1970.View ArticleGoogle Scholar
- Levy ED, Pereira-Leal JB: Evolution and dynamics of protein interactions and networks. Curr Opin Struct Biol 2008, 18(3):349–357. 10.1016/j.sbi.2008.03.003View ArticlePubMedGoogle Scholar
- Evlampiev K, Isambert H: Modeling protein network evolution under genome duplication and domain shuffling. BMC Syst Biol 2007, 1: 49. 10.1186/1752-0509-1-49PubMed CentralView ArticlePubMedGoogle Scholar
- Vázquez A, Flammini A, Maritan A, Vespignani A: Modeling of Protein Interaction Networks. Complexus 2003, 1: 38–44. 10.1159/000067642View ArticleGoogle Scholar
- Pastor-Satorras R, Smith E, Solé RV: Evolving protein interaction networks through gene duplication. J Theor Biol 2003, 222(2):199–210. 10.1016/S0022-5193(03)00028-6View ArticlePubMedGoogle Scholar
- Light S, Kraulis P, Elofsson A: Preferential attachment in the evolution of metabolic networks. BMC Genomics 2005, 6: 159. 10.1186/1471-2164-6-159PubMed CentralView ArticlePubMedGoogle Scholar
- Pereira-Leal JB, Teichmann SA: Novel specificities emerge by stepwise duplication of functional modules. Genome Res 2005, 15(4):552–559. 10.1101/gr.3102105PubMed CentralView ArticlePubMedGoogle Scholar
- Amoutzias GD, Robertson DL, Oliver SG, Bornberg-Bauer E: Convergent evolution of gene networks by single-gene duplications in higher eukaryotes. EMBO Rep 2004, 5(3):274–279. 10.1038/sj.embor.7400096PubMed CentralView ArticlePubMedGoogle Scholar
- Teichmann SA, Babu MM: Gene regulatory network growth by duplication. Nat Genet 2004, 36(5):492–496. 10.1038/ng1340View ArticlePubMedGoogle Scholar
- van Noort V, Snel B, Huynen MA: The yeast coexpression network has a small-world, scale-free architecture and can be explained by a simple model. EMBO Rep 2004, 5(3):280–284. 10.1038/sj.embor.7400090PubMed CentralView ArticlePubMedGoogle Scholar
- Eisenberg E, Levanon EY: Preferential attachment in the protein network evolution. Phys Rev Lett 2003, 91(13):138701. 10.1103/PhysRevLett.91.138701View ArticlePubMedGoogle Scholar
- Mika S, Rost B: Protein-protein interactions more conserved within species than across species. PLoS Comput Biol 2006, 2(7):e79. 10.1371/journal.pcbi.0020079PubMed CentralView ArticlePubMedGoogle Scholar
- Yu H, Luscombe NM, Lu HX, Zhu X, Xia Y, Han JDJ, Bertin N, Chung S, Vidal M, Gerstein M: Annotation transfer between genomes: protein-protein interologs and protein-DNA regulogs. Genome Res 2004, 14(6):1107–1118. 10.1101/gr.1774904PubMed CentralView ArticlePubMedGoogle Scholar
- Jose H, Vadivukarasi T, Devakumar J: Extraction of protein interaction data: a comparative analysis of methods in use. EURASIP J Bioinform Syst Biol 2007, 53096.Google Scholar
- Fawcett T: ROC Graphs: Notes and Practical Considerations for Researchers. Machine Learning 2004.Google Scholar
- Hart GT, Ramani AK, Marcotte EM: How complete are current yeast and human protein-interaction networks? Genome Biol 2006, 7(11):120. 10.1186/gb-2006-7-11-120PubMed CentralView ArticlePubMedGoogle Scholar
- Jacob F: Evolution and tinkering. Science 1977, 196(4295):1161–1166. 10.1126/science.860134View ArticlePubMedGoogle Scholar
- Kim Y, Koyutürk M, Topkara U, Grama A, Subramaniam S: Inferring functional information from domain co-evolution. Bioinformatics 2006, 22: 40–49. 10.1093/bioinformatics/bti723View ArticlePubMedGoogle Scholar
- Pearson WR, Lipman DJ: Improved tools for biological sequence comparison. Proc Natl Acad Sci USA 1988, 85(8):2444–2448. 10.1073/pnas.85.8.2444PubMed CentralView ArticlePubMedGoogle Scholar
- Altschul SF, Madden TL, Schäffer 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(17):3389–3402. 10.1093/nar/25.17.3389PubMed CentralView ArticlePubMedGoogle Scholar
- Park J, Karplus K, Barrett C, Hughey R, Haussler D, Hubbard T, Chothia C: Sequence comparisons using multiple sequences detect three times as many remote homologues as pairwise methods. J Mol Biol 1998, 284(4):1201–1210. 10.1006/jmbi.1998.2221View ArticlePubMedGoogle Scholar
- 51. Straßer W, Siegl D, Önder K, Bauer J: InSilico Proteomics System: Integration and Application of Protein and Protein-Protein Interaction Data using Microsoft .NET. Journal of Integrative Bioinformatics 2006.,3(2)Google Scholar
- Bader GD, Donaldson I, Wolting C, Ouellette BF, Pawson T, Hogue CW: BIND-The Biomolecular Interaction Network Database. Nucleic Acids Res 2001, 29: 242–245. 10.1093/nar/29.1.242PubMed CentralView ArticlePubMedGoogle Scholar
- Salwinski L, Miller CS, Smith AJ, Pettit FK, Bowie JU, Eisenberg D: The Database of Interacting Proteins: 2004 update. Nucleic Acids Res 2004, (32 Database):D449-D451. 10.1093/nar/gkh086Google Scholar
- Peri S, Navarro JD, Amanchy R, Kristiansen TZ, Jonnalagadda CK, Surendranath V, Niranjan V, Muthusamy B, Gandhi TKB, Gronborg M, Ibarrola N, Deshpande N, Shanker K, Shivashankar HN, Rashmi BP, Ramya MA, Zhao Z, Chandrika KN, Padma N, Harsha HC, Yatish AJ, Kavitha MP, Menezes M, Choudhury DR, Suresh S, Ghosh N, Saravana R, Chandran S, Krishna S, Joy M, Anand SK, Madavan V, Joseph A, Wong GW, Schiemann WP, Constantinescu SN, Huang L, Khosravi-Far R, Steen H, Tewari M, Ghaffari S, Blobe GC, Dang CV, Garcia JGN, Pevsner J, Jensen ON, Roepstorff P, Deshpande KS, Chinnaiyan AM, Hamosh A, Chakravarti A, Pandey A: Development of human protein reference database as an initial platform for approaching systems biology in humans. Genome Res 2003, 13(10):2363–2371. 10.1101/gr.1680803PubMed CentralView ArticlePubMedGoogle Scholar
- 55. Hermjakob H, Montecchi-Palazzi L, Lewington C, Mudali S, Kerrien S, Orchard S, Vingron M, Roechert B, Roepstorff P, Valencia A, Margalit H, Armstrong J, Bairoch A, Cesareni G, Sherman D, Apweiler R: IntAct: an open source molecular interaction database. Nucleic Acids Res 2004, (32 Database):D452-D455. 10.1093/nar/gkh052Google Scholar
- 56. Chatr-aryamontri A, Ceol A, Palazzi LM, Nardelli G, Schneider MV, Castagnoli L, Cesareni G: MINT: the Molecular INTeraction database. Nucleic Acids Res 2007, (35 Database):D572-D574. 10.1093/nar/gkl950Google Scholar
- 57. Güldener U, Münsterkötter M, Oesterheld M, Pagel P, Ruepp A, Mewes HW, Stümpflen V: MPact: the MIPS protein interaction resource on yeast. Nucleic Acids Res 2006, (34 Database):D436-D441. 10.1093/nar/gkj003Google Scholar
- 58. Consortium U: The universal protein resource (UniProt). Nucleic Acids Res 2008, (36 Database):D190-D195.Google Scholar
- Ben-Hur A, Noble WS: Choosing negative examples for the prediction of protein-protein interactions. BMC Bioinformatics 2006, 7(Suppl 1):S2. 10.1186/1471-2105-7-S1-S2PubMed CentralView ArticlePubMedGoogle 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.