- Research article
- Open Access
Predicting cancer involvement of genes from heterogeneous data
© Aragues et al; licensee BioMed Central Ltd. 2008
- Received: 16 June 2007
- Accepted: 27 March 2008
- Published: 27 March 2008
Systematic approaches for identifying proteins involved in different types of cancer are needed. Experimental techniques such as microarrays are being used to characterize cancer, but validating their results can be a laborious task. Computational approaches are used to prioritize between genes putatively involved in cancer, usually based on further analyzing experimental data.
We implemented a systematic method using the PIANA software that predicts cancer involvement of genes by integrating heterogeneous datasets. Specifically, we produced lists of genes likely to be involved in cancer by relying on: (i) protein-protein interactions; (ii) differential expression data; and (iii) structural and functional properties of cancer genes. The integrative approach that combines multiple sources of data obtained positive predictive values ranging from 23% (on a list of 811 genes) to 73% (on a list of 22 genes), outperforming the use of any of the data sources alone. We analyze a list of 20 cancer gene predictions, finding that most of them have been recently linked to cancer in literature.
Our approach to identifying and prioritizing candidate cancer genes can be used to produce lists of genes likely to be involved in cancer. Our results suggest that differential expression studies yielding high numbers of candidate cancer genes can be filtered using protein interaction networks.
- Cancer Type
- Cancer Gene
- Protein Interaction Network
- Evolutionary Property
- Protein Interaction Data
Tumor development results from a progressive sequence of genetic and epigenetic alterations that promote the malignant transformation of the cell by disrupting key processes involved in normal growth control and tissue homeostasis . Since complex biological networks control these processes, there are many genes that, mutated, can provide the cell with a specific aberrant capability. Alterations in three types of genes are responsible for tumorigenesis: oncogenes, tumor-suppressor genes, and stability genes . Most oncogenes are involved in controlling the rate of cell growth, while tumor suppressor genes are usually negative regulators of growth or other functions that may affect invasive and metastatic potential, such as cell adhesion and regulation of protease activity. On the other hand, stability genes control the rate of DNA mutation, and their alteration can result in mutations in oncogenes or tumor suppressor genes, thus contributing to the development of cancer .
The completion of the human genome project and the development of high-throughput experimental techniques have enabled new approaches for studying cancer. For example, gene-expression profiling using microarrays has improved the classification of some tumor types [4, 5]. Moreover, data from large-scale screenings of protein-protein interactions has been used to identify interaction subnetworks activated in cancer . Finally, genome scanning for gene copy-number alterations has detected many loci harboring candidate cancer genes . Because of these advances, efforts to catalog all of the mutational events that contribute to human cancer can now be envisioned. For example, the Cancer Genome Atlas initiative  is resequencing a substantial fraction of human genes in order to elucidate the contribution of somatic mutations to cancer development and progression. Due to the complexity of these initiatives, methods to characterize and prioritize gene candidates likely to be involved in cancer are being developed [9–12].
Protein interaction networks are a useful tool for better understanding the biology of the cell [13–15]. Moreover, the topology of the networks and the neighborhood of a given protein within the network have been used to functionally characterize proteins [16, 17]. It has also been observed that proteins related to a disease tend to have a high connectivity between them , specifically in inherited diseases [19, 20] and ataxia . Moreover, in a recent work by Barabasi and coworkers, somatic cancer genes (i.e., those that are not transmitted to descendants) were found to be more likely than other genes to encode proteins with many interaction partners (i.e., hubs) .
Gene expression profiling with DNA microarrays is a powerful approach for identifying cancer genes. Numerous studies have presented analyses of human cancer samples in which they identify gene expression signatures for different cancer types and subtypes [22–24]. In these experiments, genes are ranked according to their differential expression in the majority of cancer samples with respect to normal tissues, and genes above a predefined threshold are considered as candidate genes for the type of cancer being studied. Often, more in-depth analyses are performed to evaluate the involvement of candidate genes in the cancer, either by means of proteomics techniques , real-time polymerase chain reaction (qRT-PCR) , or literature search . However, validating the results of microarray experiments can be a long and costly effort, due to the large number of candidate genes typically involved. Often, only a handful of genes of interest are selected for experimental validation, and hundreds of others are ignored. Moreover, due to limitations in DNA microarray technology, higher differential expressions of a gene do not necessarily reflect a greater likelihood of the gene being related to cancer  and therefore, focusing only on the candidate genes with the highest differential expressions might not be the optimal procedure. Thus, there is a need for better techniques for selecting which genes will be analyzed in detail. Several procedures address the issue of selecting genes related to cancer  by further processing microarray data, either using more powerful statistics  or integrating multiple expression studies .
In order to improve the candidate gene selection process, several works have combined gene expression with other types of genomic data [32, 33]. One popular approach is gene set enrichment analysis, in which statistical tests are used to identify sets of dysregulated genes with a common biological function [34, 35]. Recently, Chinnaiyan and coworkers have combined the Molecular Concept Map and expression signatures to profile prostate cancer progression from benign epithelium to metastatic disease . In the work of Rhodes et al. , instead of relying on predefined gene annotations, they applied a human interactome to genome-wide gene expression data in cancer for identifying a potential tumor suppressor gene in the integrin signaling pathway, and then demonstrated the utility of protein-protein interaction data for identifying interaction subnetworks activated in cancer. Finally, other approaches avoid the use of high throughput data by predicting cancer genes candidates based on their sequence, structure and functional properties [9, 37].
Here, we have implemented a systematic approach for identifying genes (and gene products) involved in cancer. Our method produces lists of reliable candidate cancer genes by combining (i) a list of known cancer genes ; (ii) protein-protein interaction data ; (iii) expression information from multiple cancer studies ; and (iv) probabilities derived from structural, functional and evolutionary properties . We begin by evaluating each method separately and comparing their results. Next, we present the integrative approach and evaluate its potential for predicting cancer genes. We provide candidate cancer genes obtained as a result of this work and assess them using public repositories of biological information and literature search. We conclude by discussing potential applications of our method.
We were interested in assessing different methodologies for identifying cancer genes. Specifically, we tested the use of (i) protein interaction networks; (ii) microarray differential expression data; (iii) structural, functional and evolutionary properties of genes; and (iv) an integration of the three previous type of data. For the evaluation, we relied on a cancer gene list compiled from a variety of curated lists, cancer and sarcoma reviews, and Entrez Gene queries, followed by additional curation  (Material and Methods). We refer to genes annotated as "tumor suppressors", "oncogenes" or "stability genes" in this list as the known cancer genes. Moreover, we use the term "cancer genes" to refer to genes and proteins involved in cancer.
Predicting cancer genes based on protein interaction partners
Cancer gene enrichment of proteins according to their Cancer Linker Degree. The enrichment of cancer genes is shown for proteins with CLD ≥ 0, CLD ≥ 1 and CLD ≥ 10. The p-value of the difference between the whole data set (proteins with CLD ≥ 0) and proteins with CLD ≥ 1 and CLD ≥ 10 was calculated using the Fisher's exact test for count data (F) and the Wilcoxon signed rank test (W) on 1000 random samples.
proteins CLD ≥ 0
proteins CLD ≥ 1
p-value CLD ≥ 0 vs. CLD ≥ 1
proteins CLD ≥ 10
p-value CLD ≥ 0 vs. CLD ≥ 10
p-value CLD ≥ 1 vs. CLD ≥ 10
% of cancer genes
< 2.2 × 10-16 (F)
< 2.2 × 10-16 (W)
< 2.2 × 10-16 (F)
< 2.2 × 10-16 (W)
< 2.2 × 10-16 (F)
The CLD of a protein depends on the number of interactions that have been reported for the protein and thus, it might be influenced by how much interest has been placed on a protein by the research community. To exclude this potential bias we calculated the cancer linker degree of proteins i) using only interactions from high-throughput studies (i.e yeast two hybrid and affinity purification systems); and ii) using all interactions in PIANA except for those in the Human Protein Reference Database , which is a manually curated database of interactions extracted from the literature, with a preference towards disease related proteins. In the first case, we observed a decrease in positive predictive value (Additional file 3), while in the second scenario there was a slight increase in the positive predictive value (Additional file 4). In both cases, there is a significant enrichment of proteins with CLD ≥ 1 with respect to the average protein in the dataset (p-value of 4.8 × 10-14 and p-value < 2.2 × 10-16, respectively), concluding that the literature bias does not invalidate our initial hypothesis. Besides, similarly to previous studies [18, 42], we observed that proteins with a large number of interaction partners (i.e., hubs) are more likely to be cancer genes than proteins with few interaction partners (Additional file 5). However, using the total number of interacting partners of a protein to predict cancer genes performed worse than using the cancer linker degree: for sensitivity of ~10%, the positive predictive value was ~34%.
Predicting cancer genes based on microarray data
Cancer gene enrichment of proteins according to the number of cancer types in which they appear differentially expressed. The enrichment of cancer genes is shown for proteins differentially expressed in 1, 2 and 5 cancer types. The p-value of the difference between the different groups of proteins was calculated using the Fisher's exact test for count data (F) and the Wilcoxon signed rank test (W) on 1000 random samples.
All in dataset
1 cancer type
2 cancer types
5 cancer types
% of cancer genes
all vs. 1 < 2.2 × 10-16 (W)
all vs. 2 < 2.2 × 10-16 (W)
all vs. 5 < 2.2 × 10-16 (W)
1 vs. 2 = 2.6 × 10-11 (F)
1 vs. 2 = 2.6 × 10-11 (F)
2 vs. 5 = 2.0 × 10-13 (F)
Moreover, we studied the effect of looking at over- and under-expressed genes by their differential expression rank in a given experiment (Methods). For each differential expression study, we calculated the enrichment of cancer genes among i) the 100 most differentially expressed genes; and ii) all differentially expressed genes. None of the 24 experiments tested showed a significant increase in positive predictive value when restricting the predictions to the 100 most differentially expressed genes. These results suggest that the number of cancer types in which a gene is observed differentially expressed is a better strategy for predicting cancer genes than using its differential expression rank.
Predicting cancer genes by structural, functional and evolutionary properties
Relating the Cancer Linker Degree to differential expression and SF-Probability
Proteins with a high cancer linker degree tend to be differentially expressed in multiple cancer types
Proteins with a high cancer linker degree tend to have common functional, structural and evolutionary properties with cancer genes
We tested the correlation between the cancer linker degree (CLD) of proteins and their probabilities of being cancer genes according to their structural, functional and evolutionary properties (SF-Probabilities). We observed a significant difference between the SF-Probabilities of random proteins from the database (i.e. proteins with CLD ≥ 0) and the SF-Probabilities of proteins with interactions to cancer genes (Figure 5B). For example, we found that proteins with CLD ≥ 1 had an average SF-Probability of 0.32, which is significantly higher than for proteins with CLD ≥ 0 (SF-Probability of 0.27, p-value = 1.3 × 10-9) but significantly lower than for proteins with CLD ≥ 20 (SF-Probability of 0.51, p-value = 0.001). The lower SF-Probability of proteins with very high CLDs is explained by the few cases found with multiple interactions to known cancer genes. These results suggest that proteins with interactions to cancer genes show structural, functional and evolutionary properties similar to cancer genes.
Predicting cancer genes by integrating multiple types of data
Comparing the performances of the integrative approach and the Cancer Linker Degree method. Positive predictive values (PPV) and sensitivities are shown under nine different fixed cancer linker degrees (CLD) for a method solely based on CLD scores and an integrative approach which combines the CLD score with SF-Probability and the number of cancer types in which the gene appears differentially expressed. For all CLD thresholds above 3, the difference between the integrative approach and the CLD method alone is not significant. The p-value of the difference between the two different groups of cancer gene candidates was calculated using the Fisher's exact test.
• SF-Probability ≥ 0.3
• No. Cancer types ≥ 1
CLD ≥ 1
4.2 × 10-9
CLD ≥ 2
CLD ≥ 3
CLD ≥ 4
CLD ≥ 5
CLD ≥ 10
CLD ≥ 15
CLD ≥ 20
CLD ≥ 25
Cancer gene candidates
Cancer gene candidates. The cancer gene candidates of this table where obtained by fixing the following thresholds: (i) cancer linker degree equal of higher than 10; (ii) found differentially expressed in at least three cancer types; and (iii) probability based on structural, functional and evolutionary properties (SF-Probability) equal of higher than 0.7.
Cancer Linker degree
Number of cancer types differentially expressed
Analysis of predicted cancer genes in Table 4. Column "related to cancer" indicates whether literature  and information coming from UniProt , Reactome  and GO  indicate a strong involvement in cancer (++), somehow related to cancer (+) or not related to cancer (-). Literature references for each gene found to be involved in cancer are provided as additional file 8.
Description and Function/Pathway
Related to cancer
Cell division protein kinase 9
Regulation of progression through cell cycle
Endothelial transcription factor GATA-2
Transcriptional activator which regulates endothelin-1 gene expression
Cyclic AMP-dependent transcription factor ATF-2
Transcriptional activator which binds to the CRE, present in many viral and cellular promoters.
Essential for the control of the cell cycle at the G2/M (mitosis) transition.
Casein kinase II subunit alpha
Participates in Wnt signaling.
Peroxisome proliferator-activated receptor-binding protein
Essential for embryogenesis. Plays a role in transcriptional coactivation
Tyrosine-protein kinase CSK
Negative regulation of cell proliferation
Protein kinase A-alpha
ATP binding and protein serine/threonine kinase activity
Mediates the ubiquitination of proteins involved in cell cycle progression, signal transduction and transcription
ATP binding and protein kinase activity
Signal transducer and activator of transcription 5B
Signal transduction and activation of transcription
DNA replication licensing factor MCM7
Required for DNA replication and cell proliferation. Required for S-phase checkpoint activation upon UV-induced damage.
Mediator of RNA polymerase II transcription subunit 21
Regulation of transcription.
Macrophage-stimulating protein receptor [Precursor]
Receptor for macrophage stimulating protein (MSP). Tyrosine-protein kinase activity.
KH domain-containing, RNA-binding, signal transduction-associated protein 1
Role in G2-M progression in the cell cycle.
Tyrosine-protein kinase SYK
Positive effector of BCR-stimulated responses.
Kinase insert domain receptor
Kinase activity and receptor activity.
Nucleoside diphosphate kinase B
Major role in the synthesis of nucleoside triphosphates other than ATP.
DNA-directed RNA polymerase II 140 kDa polypeptide
DNA-dependent RNA polymerase catalyzes the transcription of DNA into RNA.
Serum response factor
SRF is a transcription factor that binds to the serum response element (SRE)
We analyzed the use of three different criteria for predicting cancer gene candidates and concluded that: (i) the number of interaction partners of a protein that have been previously annotated as cancer gene (i.e. the cancer linker degree) is a good indicator of the likelihood of the protein to be involved in cancer; (ii) using differences in gene expression between normal tissue and cancer identifies many known cancer genes, but many non cancer genes as well; and (iii) probabilities based on structural, functional and evolutionary properties of known cancer genes (i.e. SF-Probabilities) are useful for filtering false positives from other cancer gene prediction methods. Moreover, we implemented and evaluated a method that integrates these criteria to produce reliable lists of cancer gene candidates, obtaining a positive predictive value of 73% when using very restrictive thresholds. Finally, we provided lists of cancer gene candidates and analyzed them using literature sources and information from public repositories, showing that our predictions are reliable.
Most methods used for predicting or prioritizing cancer gene candidates are biased towards genes that are well annotated and/or familiar to the researcher. This leaves unexplored many potential cancer gene candidates. However, high throughput genomic and proteomic work has now yielded relatively unbiased, although noisy, genome- and proteome-wide data sets. For example, expression studies produce large lists of over- and under-expressed genes, which are then prioritized by their differential expression rank, usually with help of a limited number of literature searches. Our integrative approach to finding cancer gene candidates can be used to obtain unbiased lists of cancer gene candidates by using the cancer linker degree of proteins to filter expression studies. We observed that the low positive predictive value obtained when using differential expression data alone (around 15% for most cancer types in our study) shows a four-fold increase when combined with protein-protein interaction data. We expect that further experimental study of our proposed cancer gene candidates will find useful the methodology presented in this work.
Separately, each of the criterion presented here for cancer gene candidate prediction has its limitations. First, methods based on protein interaction networks are limited by the fact that many cancers are the result of perturbations in the regulation of genes, which is not captured by protein-protein interaction data. Second, differential expression based methods have the drawback that many differentially expressed genes are not a cause for the cancer but rather a consequence of it. Besides, we are mapping expression levels of mRNA onto a network of protein interactions. However, it is known that the mRNA expression levels do not always match the protein expression levels . Finally, methods based on structural, functional and evolutionary properties are very dependent on existing functional annotations (e.g. available GO information for a given protein) and their predictions are more stochastic than based on biological observations. These limitations could be avoided by the use of types of information such as gene regulatory networks  and gene copy-number alterations . Moreover, recently developed experimental techniques promise an increase in the amount and types of data available , including protein post-translational modifications , tissue localization  and protein expression in specific cancers . Finally, the integrative approach is constrained by the limitations of each independent method. However, depending on the context of application, these limitations can be avoided by ignoring irrelevant data: for example, SF-Probabilities should not be used when searching for cancer genes of unknown function.
Our reported performance results on the use of SF-Probabilities differ markedly from the evaluation presented by Lopez-Bigas and coworkers . We attribute this difference to two factors: (i) we used a more extensive set of known cancer genes; (ii) we used different evaluation metrics and methods: for example, Lopez-Bigas and coworkers used a balanced dataset to evaluate their method, whereas we considered as non-cancer gene any gene that was not a known cancer gene. We believe that the performance metrics and evaluation method used in this work are more representative of predictions done on the full human genome.
The methods presented here were evaluated by comparing their cancer gene predictions with a curated list of oncogenes, tumor suppressors and stability genes . This list of known cancer genes attempts to be as comprehensive as possible, but two possible biases arise from it: (i) not all methods cover the space of cancer genes to the same extent (e.g. the model used to calculate SF-Probabilities was trained on genes for which mutations have been causally implicated in cancer); and (ii) the method based on protein interaction networks heavily relies on the initial set of seed cancer genes and thus, genes isolated in the cancer network will never be pinpointed. An alternative approach to seeding our method with a list of known cancer genes is one where the seeds for building the protein interaction network are cancer-related proteins obtained with low-throughput experimental methods [57, 58]. This would remove the bias introduced by the input list of known cancer genes.
We showed that the integration of multiple sources of data is more reliable for predicting cancer genes than the use of one single criterion. For example, differential expression studies could benefit from the use of protein-protein interaction data to further validate their results: in the best case scenario, combining the cancer linker degree of a protein with differential expression data increased from 17% to 73% the fraction of known cancer genes within the cancer gene candidates. In conclusion, systems capable of integrating all available sources of data are fundamental to the discovery of proteins involved in cancer.
Known cancer genes
We downloaded cancer genes from the Memorial Sloan Kettering computational biology website CancerGenes  as of January 2007. We collected a set of known cancer genes by querying the website for "oncogene", "tumor suppressor" and "stability". This list comprised 1,256 cancer genes, in particular 385 oncogenes, 471 tumor suppressors and 494 stability genes (several genes belonged to more than one category).
Protein Interaction Data
We used PIANA  to integrate human protein interaction data from DIP 2007.02.19 , MIPS 2007.04.03 , HPRD v6.01 , BIND 2007.04.03 , IntAct 2007.04.23 , BioGrid v2.026  and MINT 2007.04.05 . The integration of different sources of interactions into a single database allowed us to work with an extensive set of 110,457 human interactions between 36,900 different protein sequences. This set of human interaction data includes 24,812 interactions from yeast two-hybrid assays, 13,256 interactions from immunoprecipitation methods and 11,174 interactions from affinity chromatography methods. HPRD, a database manually curated from literature sources contained 38,762 interactions.
PIANA represents the protein interaction data as a network where the nodes are proteins and the edges interactions between the proteins. In such a network, a set of proteins linked to protein pj (ie, physically interacting with pj) is named "partners of pj". PIANA builds the network by retrieving direct interaction partners for an initial set of seed proteins (i.e. the proteins of interest).
We manually searched for gene expression studies between normal tissue and cancer in Oncomine , a cancer profiling database. We downloaded lists of over- and under-expressed genes from a total of 24 Oncomine studies, corresponding to 12 different cancer types (see additional file 6 for the list of experiments, the cancer type category assigned to them, and the total number of over- and under-expressed genes in each experiment). A gene was considered to have a significant differential expression if its Q value was lower than 0.05. Q values are assigned in Oncomine by correcting for multiple hypothesis testing the p-values calculated using Student's t-test for two-class differential expression analyses. A detailed description of the normalization process and statistical tests used in Oncomine can be found in [36, 39].
Probabilities of being cancer-gene based on structural and functional properties
We used the probabilities of being a cancer gene calculated in  for human genes. These probabilities were obtained using a Bayesian classification model that scored human genes for their likelihood of involvement in cancer according to structural, functional and evolutionary properties. Specifically, Lopez-Bigas and coworkers  relied on GO annotations  and sequence properties such as the extent of conservation, paralogy, and the lengths of proteins and genes. We refer to these estimated probabilities as SF-Probabilities. 12,194 human genes had an associated SF-Probability, 240 of which had been used to train the Bayesian model. 706 human genes had an SF-Probability higher than 0.95, and the SF-Probability was lower than 0.1 for 6288 human genes. Finally, 758 genes did not have an associated protein sequence in PIANA and thus, were not used in this work.
Genes, proteins and identifiers
We used PIANA  to map expression data and SF-Probabilities onto the interaction network, in particular gene symbols coming from Oncomine expression studies and Ensembl identifiers coming from . Throughout the text, we use the term 'cancer gene' to refer to any gene or protein involved in cancer.
Evaluating the use of protein interaction networks to predict cancer genes
The cancer protein interaction network was built using PIANA  by setting the list of known cancer genes as seeds (see "protein interaction data", Material and Methods). In this network, we define the cancer linker degree (CLD) of a protein as the number of cancer genes to which it is directly connected (Figure 1). The CLD was calculated for each protein and proteins were binned by their CLDs. In this context, and given a CLD threshold of N, positives are proteins with CLD ≥ N. True positives are known cancer genes among positives. False negatives are known cancer genes whose CLD is lower than N. The positive predictive value is defined as the ratio between true positives and positives. Sensitivity is the ratio between true positives and the sum of false negatives and true positives. Positive predictive values and sensitivities are shown in Figure 2 for CLD thresholds with at least 5 positives.
Evaluating the use of differential expression data to predict cancer genes
We calculated how many over- or under-expressed genes were known cancer genes for each cancer type described on Additional file 6. Moreover, we tested how many genes differentially expressed in at least 1–5 cancer types were known cancer genes. In this context, any differentially expressed gene is considered a positive. Among positives, we define as true positives those that are known cancer genes. False negatives are known cancer genes not found differentially expressed. Besides, we evaluated the prediction of cancer genes based on the differential expression rank of the cancer gene candidates in the lists of over- and under-expressed genes from Oncomine . In particular, we analyzed the enrichment of cancer genes among the 50 most differentially expressed genes in the lists of over- and under-expressed genes, and compared it to the enrichment of cancer genes among all differentially expressed genes.
Evaluating the use of structural, functional and evolutionary properties to predict cancer genes
At any given SF-Probability threshold, positives are proteins with a SF-Probability above or equal to that threshold. Among positives, true positives are those that are known cancer genes. False negatives are known cancer genes not found above the SF-Probability threshold. Genes used for training the model in  were discarded for the evaluation.
Protein functions, pathways and literature
We manually analyzed cancer gene predictions from Table 4 by examining (i) the protein function and description as defined in UniProt ; (ii) the pathways in which the protein participated according to Reactome ; (iii) the molecular function and biological process as classified in the Gene Ontology (GO) ; and (iv) published articles retrieved using iHop .
The assessment on whether two binomial samples of observations are significantly different was calculated using Fisher's exact test on a 2 × 2 contingency table comparing the number of cancer genes and non-cancer genes between two groups (e.g. CLD ≥ 10 versus CLD ≥ 1). The assessment on whether a distribution of averages on the number of cancer genes calculated on random samples is significantly different from a given ratio of cancer genes was calculated using the Wilcoxon signed rank test (e.g. ratio of cancer genes found on the 5537 proteins with CLD ≥ 1 versus 1000 averages extracted from random samples of size 5537). The assessment on whether two non-Gaussian samples of observations (SF-Probabilities or number of cancer types grouped by proteins with the same CLD) come from the same distribution was calculated using the Mann-Whitney U two-sided test. Differences in the observations were considered significant for p-values lower than 0.05. All tests were performed using the implementation provided by R .
We provide the complete list of human genes with the corresponding cancer gene prediction scores according to each type of data at http://sbi.imim.es/piana/scored_genes.tab.txt.
We thank N. Lopez-Bigas for providing the SF-Probability data and helpful comments on the manuscript. We thank Carlos Rodriguez and all members of cbio at mskcc for helpful discussions and comments, especially Emek Demir, Robert Hoffmann, Doron Betel and Nikolaus Schultz. RA is supported by a grant from the Spanish Ministerio de Ciencia y Tecnología (MCyT, BIO2002-03609). The work has been supported by grants from the Spanish Ministerio de Educación y Ciencia (MEC, BIO02005-00533) and from the Spanish Ministerio de Ciencia y Tecnologia (PROFIT PSE-010000-2007-1 and FIT-350300-2006-40/41/42).
- Hanahan D, Weinberg RA: The hallmarks of cancer. Cell 2000, 100: 57–70. 10.1016/S0092-8674(00)81683-9View ArticlePubMedGoogle Scholar
- Vogelstein B, Kinzler KW: Cancer genes and the pathways they control. Nat Med 2004, 10: 789–799. 10.1038/nm1087View ArticlePubMedGoogle Scholar
- Bielas JH, Loeb KR, Rubin BP, True LD, Loeb LA: Human cancers express a mutator phenotype. Proceedings of the National Academy of Sciences of the United States of America 2006, 103: 18238–18242. 10.1073/pnas.0607057103PubMed CentralView ArticlePubMedGoogle Scholar
- Segal E, Friedman N, Koller D, Regev A: A module map showing conditional activity of expression modules in cancer. Nat Genet 2004, 36: 1090–1098. 10.1038/ng1434View ArticlePubMedGoogle Scholar
- Fan C, Oh DS, Wessels L, Weigelt B, Nuyten DS, Nobel AB, van't Veer LJ, Perou CM: Concordance among gene-expression-based predictors for breast cancer. The New England journal of medicine 2006, 355: 560–569. 10.1056/NEJMoa052933View ArticlePubMedGoogle Scholar
- Rhodes DR, Tomlins SA, Varambally S, Mahavisno V, Barrette T, Kalyana-Sundaram S, Ghosh D, Pandey A, Chinnaiyan AM: Probabilistic model of the human protein-protein interaction network. Nat Biotechnol 2005, 23: 951–959. 10.1038/nbt1103View ArticlePubMedGoogle Scholar
- Lucito R, Healy J, Alexander J, Reiner A, Esposito D, Chi M, Rodgers L, Brady A, Sebat J, Troge J, West JA, Rostan S, Nguyen KC, Powers S, Ye KQ, Olshen A, Venkatraman E, Norton L, Wigler M: Representational oligonucleotide microarray analysis: a high-resolution method to detect genome copy number variation. Genome Res 2003, 13: 2291–2305. 10.1101/gr.1349003PubMed CentralView ArticlePubMedGoogle Scholar
- Futreal PA, Coin L, Marshall M, Down T, Hubbard T, Wooster R, Rahman N, Stratton MR: A census of human cancer genes. Nat Rev Cancer 2004, 4: 177–183. 10.1038/nrc1299PubMed CentralView ArticlePubMedGoogle Scholar
- Hu P, Bader G, Wigle DA, Emili A: Computational prediction of cancer-gene function. Nat Rev Cancer 2007, 7: 23–34. 10.1038/nrc2036View ArticlePubMedGoogle Scholar
- Higgins ME, Claremont M, Major JE, Sander C, Lash AE: CancerGenes: a gene selection resource for cancer genome projects. Nucleic Acids Res 2007, 35: D721–6. 10.1093/nar/gkl811PubMed CentralView ArticlePubMedGoogle Scholar
- Nguyen DX, Massague J: Genetic determinants of cancer metastasis. Nature reviews 2007, 8: 341–352. 10.1038/nrg2101View ArticlePubMedGoogle Scholar
- Barabasi AL, Oltvai ZN: Network biology: understanding the cell's functional organization. Nat Rev Genet 2004, 5: 101–113. 10.1038/nrg1272View ArticlePubMedGoogle Scholar
- Gavin AC, Bösche M, Krause R, Grandi P, Marzioch M, Bauer A, Schultz J, Rick JM, Michon AM, Cruciat CM, Remor M, Höfert C, Schelder M, Brajenovic M, Ruffner H, Merino A, Klein K, Hudak M, Dickson D, Rudi T, Gnau V, Bauch A, Bastuck S, Huhse B, Leutwein C, Heurtier MA, Copley RR, Edelmann A, Querfurth E, Rybin V, Drewes G, Raida M, Bouwmeester T, Bork P, Seraphin B, Kuster B, Neubauer G, Superti-Furga G: Functional organization of the yeast proteome by systematic analysis of protein complexes. Nature 2002, 415: 141–147. 10.1038/415141aView ArticlePubMedGoogle Scholar
- Gavin AC, Aloy P, Grandi P, Krause R, Boesche M, Marzioch M, Rau C, Jensen LJ, Bastuck S, Dumpelfeld B, Edelmann A, Heurtier MA, Hoffman V, Hoefert C, Klein K, Hudak M, Michon AM, Schelder M, Schirle M, Remor M, Rudi T, Hooper S, Bauer A, Bouwmeester T, Casari G, Drewes G, Neubauer G, Rick JM, Kuster B, Bork P, Russell RB, Superti-Furga G: Proteome survey reveals modularity of the yeast cell machinery. Nature 2006, 440: 631–636. 10.1038/nature04532View ArticlePubMedGoogle Scholar
- Sharan R, Ideker T: Modeling cellular machinery through biological network comparison. Nat Biotechnol 2006, 24: 427–433. 10.1038/nbt1196View ArticlePubMedGoogle Scholar
- Schwikowski B, Uetz P, Fields S: A network of protein-protein interactions in yeast. Nat Biotechnol 2000, 18: 1257–1261. 10.1038/82360View ArticlePubMedGoogle Scholar
- Goh KI, Cusick ME, Valle D, Childs B, Vidal M, Barabasi AL: The human disease network. Proc Natl Acad Sci U S A 2007, 104: 8685–8690. 10.1073/pnas.0701361104PubMed CentralView ArticlePubMedGoogle Scholar
- Gandhi TK, Zhong J, Mathivanan S, Karthick L, Chandrika KN, Mohan SS, Sharma S, Pinkert S, Nagaraju S, Periaswamy B, Mishra G, Nandakumar K, Shen B, Deshpande N, Nayak R, Sarker M, Boeke JD, Parmigiani G, Schultz J, Bader JS, Pandey A: Analysis of the human protein interactome and comparison with yeast, worm and fly interaction datasets. Nat Genet 2006, 38: 285–293. 10.1038/ng1747View ArticlePubMedGoogle Scholar
- Lage K, Karlberg EO, Storling ZM, Olason PI, Pedersen AG, Rigina O, Hinsby AM, Tumer Z, Pociot F, Tommerup N, Moreau Y, Brunak S: A human phenome-interactome network of protein complexes implicated in genetic disorders. Nat Biotechnol 2007, 25: 309–316. 10.1038/nbt1295View ArticlePubMedGoogle Scholar
- Lim J, Hao T, Shaw C, Patel AJ, Szabo G, Rual JF, Fisk CJ, Li N, Smolyar A, Hill DE, Barabasi AL, Vidal M, Zoghbi HY: A protein-protein interaction network for human inherited ataxias and disorders of Purkinje cell degeneration. Cell 2006, 125: 801–814. 10.1016/j.cell.2006.03.032View ArticlePubMedGoogle Scholar
- Dhanasekaran SM, Barrette TR, Ghosh D, Shah R, Varambally S, Kurachi K, Pienta KJ, Rubin MA, Chinnaiyan AM: Delineation of prognostic biomarkers in prostate cancer. Nature 2001, 412: 822–826. 10.1038/35090585View ArticlePubMedGoogle Scholar
- Notterman DA, Alon U, Sierk AJ, Levine AJ: Transcriptional gene expression profiles of colorectal adenoma, adenocarcinoma, and normal tissue examined by oligonucleotide arrays. Cancer Res 2001, 61: 3124–3130.PubMedGoogle Scholar
- Welsh JB, Zarrinkar PP, Sapinoso LM, Kern SG, Behling CA, Monk BJ, Lockhart DJ, Burger RA, Hampton GM: Analysis of gene expression profiles in normal and neoplastic ovarian tissue samples identifies candidate molecular markers of epithelial ovarian cancer. Proc Natl Acad Sci U S A 2001, 98: 1176–1181. 10.1073/pnas.98.3.1176PubMed CentralView ArticlePubMedGoogle Scholar
- Cho WC: Contribution of oncoproteomics to cancer biomarker discovery. Molecular cancer 2007, 6: 25. 10.1186/1476-4598-6-25PubMed CentralView ArticlePubMedGoogle Scholar
- Kuo WP, Liu F, Trimarchi J, Punzo C, Lombardi M, Sarang J, Whipple ME, Maysuria M, Serikawa K, Lee SY, McCrann D, Kang J, Shearstone JR, Burke J, Park DJ, Wang X, Rector TL, Ricciardi-Castagnoli P, Perrin S, Choi S, Bumgarner R, Kim JH, Short GF 3rd, Freeman MW, Seed B, Jensen R, Church GM, Hovig E, Cepko CL, Park P, Ohno-Machado L, Jenssen TK: A sequence-oriented comparison of gene expression measurements across different hybridization-based technologies. Nature biotechnology 2006, 24: 832–840. 10.1038/nbt1217View ArticlePubMedGoogle Scholar
- Jensen LJ, Saric J, Bork P: Literature mining for the biologist: from information retrieval to biological discovery. Nature reviews 2006, 7: 119–129. 10.1038/nrg1768View ArticlePubMedGoogle Scholar
- Draghici S, Khatri P, Eklund AC, Szallasi Z: Reliability and reproducibility issues in DNA microarray measurements. Trends Genet 2006, 22(2):101–109. 10.1016/j.tig.2005.12.005PubMed CentralView ArticlePubMedGoogle Scholar
- Allison DB, Cui X, Page GP, Sabripour M: Microarray data analysis: from disarray to consolidation and consensus. Nat Rev Genet 2006, 7: 55–65. 10.1038/nrg1749View ArticlePubMedGoogle Scholar
- Mehta T, Tanik M, Allison DB: Towards sound epistemological foundations of statistical methods for high-dimensional biology. Nat Genet 2004, 36: 943–947. 10.1038/ng1422View ArticlePubMedGoogle Scholar
- Rhodes DR, Yu J, Shanker K, Deshpande N, Varambally R, Ghosh D, Barrette T, Pandey A, Chinnaiyan AM: Large-scale meta-analysis of cancer microarray data identifies common transcriptional profiles of neoplastic transformation and progression. Proc Natl Acad Sci U S A 2004, 101: 9309–9314. 10.1073/pnas.0401994101PubMed CentralView ArticlePubMedGoogle Scholar
- Rhodes DR, Chinnaiyan AM: Integrative analysis of the cancer transcriptome. Nat Genet 2005, 37 Suppl: S31–7. 10.1038/ng1570View ArticlePubMedGoogle Scholar
- Mathew JP, Taylor BS, Bader GD, Pyarajan S, Antoniotti M, Chinnaiyan AM, Sander C, Burakoff SJ, Mishra B: From bytes to bedside: data integration and computational biology for translational cancer research. PLoS computational biology 2007, 3: e12. 10.1371/journal.pcbi.0030012PubMed CentralView ArticlePubMedGoogle Scholar
- Mootha VK, Lindgren CM, Eriksson KF, Subramanian A, Sihag S, Lehar J, Puigserver P, Carlsson E, Ridderstrale M, Laurila E, Houstis N, Daly MJ, Patterson N, Mesirov JP, Golub TR, Tamayo P, Spiegelman B, Lander ES, Hirschhorn JN, Altshuler D, Groop LC: PGC-1alpha-responsive genes involved in oxidative phosphorylation are coordinately downregulated in human diabetes. Nat Genet 2003, 34: 267–273. 10.1038/ng1180View ArticlePubMedGoogle Scholar
- Subramanian A, Tamayo P, Mootha VK, Mukherjee S, Ebert BL, Gillette MA, Paulovich A, Pomeroy SL, Golub TR, Lander ES, Mesirov JP: Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proc Natl Acad Sci U S A 2005, 102: 15545–15550. 10.1073/pnas.0506580102PubMed CentralView ArticlePubMedGoogle Scholar
- Tomlins SA, Mehra R, Rhodes DR, Cao X, Wang L, Dhanasekaran SM, Kalyana-Sundaram S, Wei JT, Rubin MA, Pienta KJ, Shah RB, Chinnaiyan AM: Integrative molecular concept modeling of prostate cancer progression. Nat Genet 2007, 39: 41–51. 10.1038/ng1935View ArticlePubMedGoogle Scholar
- Furney SJ, Higgins DG, Ouzounis CA, Lopez-Bigas N: Structural and functional properties of genes involved in human cancer. BMC Genomics 2006, 7: 3. 10.1186/1471-2164-7-3PubMed CentralView ArticlePubMedGoogle Scholar
- Aragues R, Jaeggi D, Oliva B: PIANA: protein interactions and network analysis. Bioinformatics 2006, 22: 1015–1017. 10.1093/bioinformatics/btl072View ArticlePubMedGoogle Scholar
- Rhodes DR, Yu J, Shanker K, Deshpande N, Varambally R, Ghosh D, Barrette T, Pandey A, Chinnaiyan AM: ONCOMINE: a cancer microarray database and integrated data-mining platform. Neoplasia 2004, 6: 1–6.PubMed CentralView ArticlePubMedGoogle Scholar
- Aouacheria A, Navratil V, Wen W, Jiang M, Mouchiroud D, Gautier C, Gouy M, Zhang M: In silico whole-genome scanning of cancer-associated nonsynonymous SNPs and molecular characterization of a dynein light chain tumour variant. Oncogene 2005, 24: 6133–6142. 10.1038/sj.onc.1208745View ArticlePubMedGoogle Scholar
- Peri S, Navarro JD, Kristiansen TZ, Amanchy R, Surendranath V, Muthusamy B, Gandhi TK, Chandrika KN, Deshpande N, Suresh S, Rashmi BP, Shanker K, Padma N, Niranjan V, Harsha HC, Talreja N, Vrushabendra BM, Ramya MA, Yatish AJ, Joy M, Shivashankar HN, Kavitha MP, Menezes M, Choudhury DR, Ghosh N, Saravana R, Chandran S, Mohan S, Jonnalagadda CK, Prasad CK, Kumar-Sinha C, Deshpande KS, Pandey A: Human protein reference database as a discovery resource for proteomics. Nucleic Acids Res 2004, 32: D497–501. 10.1093/nar/gkh070PubMed CentralView ArticlePubMedGoogle Scholar
- Jonsson PF, Bates PA: Global topological features of cancer proteins in the human interactome. Bioinformatics (Oxford, England) 2006, 22: 2291–2297. 10.1093/bioinformatics/btl390View ArticleGoogle Scholar
- Hoffmann R, Valencia A: A gene network for navigating the literature. Nature genetics 2004, 36: 664. 10.1038/ng0704-664View ArticlePubMedGoogle Scholar
- The Universal Protein Resource (UniProt) Nucleic Acids Res 2007, 35: D193–7. 10.1093/nar/gkl929Google Scholar
- Joshi-Tope G, Gillespie M, Vastrik I, D'Eustachio P, Schmidt E, de Bono B, Jassal B, Gopinath GR, Wu GR, Matthews L, Lewis S, Birney E, Stein L: Reactome: a knowledgebase of biological pathways. Nucleic Acids Res 2005, 33: D428–32. 10.1093/nar/gki072PubMed CentralView ArticlePubMedGoogle Scholar
- Harris MA, Clark J, Ireland A, Lomax J, Ashburner M, Foulger R, Eilbeck K, Lewis S, Marshall B, Mungall C, Richter J, Rubin GM, Blake JA, Bult C, Dolan M, Drabkin H, Eppig JT, Hill DP, Ni L, Ringwald M, Balakrishnan R, Cherry JM, Christie KR, Costanzo MC, Dwight SS, Engel S, Fisk DG, Hirschman JE, Hong EL, Nash RS, Sethuraman A, Theesfeld CL, Botstein D, Dolinski K, Feierbach B, Berardini T, Mundodi S, Rhee SY, Apweiler R, Barrell D, Camon E, Dimmer E, Lee V, Chisholm R, Gaudet P, Kibbe W, Kishore R, Schwarz EM, Sternberg P, Gwinn M, Hannick L, Wortman J, Berriman M, Wood V, de la Cruz N, Tonellato P, Jaiswal P, Seigfried T, White R: The Gene Ontology (GO) database and informatics resource. Nucleic Acids Res 2004, 32: D258–61. 10.1093/nar/gkh066View ArticlePubMedGoogle Scholar
- Hong JJ, Yankee TM, Harrison ML, Geahlen RL: Regulation of signaling in B cells through the phosphorylation of Syk on linker region tyrosines. A mechanism for negative signaling by the Lyn tyrosine kinase. The Journal of biological chemistry 2002, 277: 31703–31714. 10.1074/jbc.M201362200View ArticlePubMedGoogle Scholar
- Kunze E, Wendt M, Schlott T: Promoter hypermethylation of the 14–3-3 sigma, SYK and CAGE-1 genes is related to the various phenotypes of urinary bladder carcinomas and associated with progression of transitional cell carcinomas. International journal of molecular medicine 2006, 18: 547–557.PubMedGoogle Scholar
- Yuan Y, Wang J, Li J, Wang L, Li M, Yang Z, Zhang C, Dai JL: Frequent epigenetic inactivation of spleen tyrosine kinase gene in human hepatocellular carcinoma. Clinical cancer research 2006, 12: 6687–6695. 10.1158/1078-0432.CCR-06-0921PubMed CentralView ArticlePubMedGoogle Scholar
- Welm AL, Sneddon JB, Taylor C, Nuyten DS, van de Vijver MJ, Hasegawa BH, Bishop JM: The macrophage-stimulating protein pathway promotes metastasis in a mouse model for breast cancer and predicts poor prognosis in humans. Proceedings of the National Academy of Sciences of the United States of America 2007, 104: 7570–7575. 10.1073/pnas.0702095104PubMed CentralView ArticlePubMedGoogle Scholar
- Lee HJ, Yun CH, Lim SH, Kim BC, Baik KG, Kim JM, Kim WH, Kim SJ: SRF is a nuclear repressor of Smad3-mediated TGF-beta signaling. Oncogene 2007, 26: 173–185. 10.1038/sj.onc.1209774View ArticlePubMedGoogle Scholar
- Tian Q, Stepaniants SB, Mao M, Weng L, Feetham MC, Doyle MJ, Yi EC, Dai H, Thorsson V, Eng J, Goodlett D, Berger JP, Gunter B, Linseley PS, Stoughton RB, Aebersold R, Collins SJ, Hanlon WA, Hood LE: Integrated genomic and proteomic analyses of gene expression in Mammalian cells. Mol Cell Proteomics 2004, 3: 960–969. 10.1074/mcp.M400055-MCP200View ArticlePubMedGoogle Scholar
- Davidson EH, Rast JP, Oliveri P, Ransick A, Calestani C, Yuh CH, Minokawa T, Amore G, Hinman V, Arenas-Mena C, Otim O, Brown CT, Livi CB, Lee PY, Revilla R, Rust AG, Pan Z, Schilstra MJ, Clarke PJ, Arnone MI, Rowen L, Cameron RA, McClay DR, Hood L, Bolouri H: A genomic regulatory network for development. Science 2002, 295: 1669–1678. 10.1126/science.1069883View ArticlePubMedGoogle Scholar
- Ptacek J, Devgan G, Michaud G, Zhu H, Zhu X, Fasolo J, Guo H, Jona G, Breitkreutz A, Sopko R, McCartney RR, Schmidt MC, Rachidi N, Lee SJ, Mah AS, Meng L, Stark MJ, Stern DF, De Virgilio C, Tyers M, Andrews B, Gerstein M, Schweitzer B, Predki PF, Snyder M: Global analysis of protein phosphorylation in yeast. Nature 2005, 438: 679–684. 10.1038/nature04187View ArticlePubMedGoogle Scholar
- Huh WK, Falvo JV, Gerke LC, Carroll AS, Howson RW, Weissman JS, O'Shea EK: Global analysis of protein localization in budding yeast. Nature 2003, 425: 686–691. 10.1038/nature02026View ArticlePubMedGoogle Scholar
- Varambally S, Yu J, Laxman B, Rhodes DR, Mehra R, Tomlins SA, Shah RB, Chandran U, Monzon FA, Becich MJ, Wei JT, Pienta KJ, Ghosh D, Rubin MA, Chinnaiyan AM: Integrative genomic and proteomic analysis of prostate cancer reveals signatures of metastatic progression. Cancer Cell 2005, 8: 393–406. 10.1016/j.ccr.2005.10.001View ArticlePubMedGoogle Scholar
- Espana L, Martin B, Aragues R, Chiva C, Oliva B, Andreu D, Sierra A: Bcl-x(L)-mediated changes in metabolic pathways of breast cancer cells: from survival in the blood stream to organ-specific metastasis. Am J Pathol 2005, 167: 1125–1137.PubMed CentralView ArticlePubMedGoogle Scholar
- Mendez O, Martin B, Sanz R, Aragues R, Moreno V, Oliva B, Stresing V, Sierra A: Underexpression of transcriptional regulators is common in metastatic breast cancer cells overexpressing Bcl-xL. Carcinogenesis 2006, 27: 1169–1179. 10.1093/carcin/bgi363View 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: D449–51. 10.1093/nar/gkh086PubMed CentralView ArticlePubMedGoogle Scholar
- Pagel P, Kovac S, Oesterheld M, Brauner B, Dunger-Kaltenbach I, Frishman G, Montrone C, Mark P, Stumpflen V, Mewes HW, Ruepp A, Frishman D: The MIPS mammalian protein-protein interaction database. Bioinformatics 2005, 21: 832–834. 10.1093/bioinformatics/bti115View ArticlePubMedGoogle Scholar
- Alfarano C, Andrade CE, Anthony K, Bahroos N, Bajec M, Bantoft K, Betel D, Bobechko B, Boutilier K, Burgess E, Buzadzija K, Cavero R, D'Abreo C, Donaldson I, Dorairajoo D, Dumontier MJ, Dumontier MR, Earles V, Farrall R, Feldman H, Garderman E, Gong Y, Gonzaga R, Grytsan V, Gryz E, Gu V, Haldorsen E, Halupa A, Haw R, Hrvojic A, Hurrell L, Isserlin R, Jack F, Juma F, Khan A, Kon T, Konopinsky S, Le V, Lee E, Ling S, Magidin M, Moniakis J, Montojo J, Moore S, Muskat B, Ng I, Paraiso JP, Parker B, Pintilie G, Pirone R, Salama JJ, Sgro S, Shan T, Shu Y, Siew J, Skinner D, Snyder K, Stasiuk R, Strumpf D, Tuekam B, Tao S, Wang Z, White M, Willis R, Wolting C, Wong S, Wrong A, Xin C, Yao R, Yates B, Zhang S, Zheng K, Pawson T, Ouellette BF, Hogue CW: The Biomolecular Interaction Network Database and related tools 2005 update. Nucleic Acids Res 2005, 33: D418–24. 10.1093/nar/gki051PubMed CentralView ArticlePubMedGoogle Scholar
- Kerrien S, Alam-Faruque Y, Aranda B, Bancarz I, Bridge A, Derow C, Dimmer E, Feuermann M, Friedrichsen A, Huntley R, Kohler C, Khadake J, Leroy C, Liban A, Lieftink C, Montecchi-Palazzi L, Orchard S, Risse J, Robbe K, Roechert B, Thorneycroft D, Zhang Y, Apweiler R, Hermjakob H: IntAct--open source resource for molecular interaction data. Nucleic Acids Res 2007, 35: D561–5. 10.1093/nar/gkl958PubMed CentralView ArticlePubMedGoogle Scholar
- Stark C, Breitkreutz BJ, Reguly T, Boucher L, Breitkreutz A, Tyers M: BioGRID: a general repository for interaction datasets. Nucleic Acids Res 2006, 34: D535–9. 10.1093/nar/gkj109PubMed CentralView ArticlePubMedGoogle Scholar
- 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: D572–4. 10.1093/nar/gkl950PubMed CentralView ArticlePubMedGoogle Scholar
- Ihaka R, Gentleman R: R: A language for data analysis and graphics. Journal of Computational and Graphical Statistics 1996, 5(3):299–314. [http://www.r-project.org/doc/bib/R-other_bib.html#R:Ihaka+Gentleman:1996] 10.2307/1390807Google 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.