Patterns of human gene expression variance show strong associations with signaling network hierarchy
© Komurov and Ram; licensee BioMed Central Ltd. 2010
Received: 2 April 2010
Accepted: 12 November 2010
Published: 12 November 2010
Understanding organizational principles of cellular networks is one of the central goals of systems biology. Although much has been learnt about gene expression programs under specific conditions, global patterns of expressional variation (EV) of genes and their relationship to cellular functions and physiological responses is poorly understood.
To understand global principles of relationship between transcriptional regulation of human genes and their functions, we have leveraged large-scale datasets of human gene expression measurements across a wide spectrum of cell conditions. We report that human genes are highly diverse in terms of their EV; while some genes have highly variable expression pattern, some seem to be relatively ubiquitously expressed across a wide range of conditions. The wide spectrum of gene EV strongly correlates with the positioning of proteins within the signaling network hierarchy, such that, secreted extracellular receptor ligands and membrane receptors have the highest EV, and intracellular signaling proteins have the lowest EV in the genome. Our analysis shows that this pattern of EV reflects functional centrality: proteins with highly specific signaling functions are modulated more frequently than those with highly central functions in the network, which is also consistent with previous studies on tissue-specific gene expression. Interestingly, these patterns of EV along the signaling network hierarchy have significant correlations with promoter architectures of respective genes.
Our analyses suggest a generic systems level mechanism of regulation of the cellular signaling network at the transcriptional level.
Gene expression changes in the cell allow for reprogramming of cellular behavior depending on the extracellular conditions. Global gene expression profiling of cells has become a routine procedure in biology, and extensive work has been done in the recent years studying gene expression programs under various conditions [1–4]. In addition, many aspects of gene expression behavior at the DNA and chromatin level have also been identified [5–9]. Although these studies yielded much insight into the regulation of gene expression under the specific conditions studied, we do not have a clear understanding of global patterns in gene expression regulation in human cells in response to extracellular stimuli. Some notable studies addressing functional aspects of gene expression regulation at a systems level have been performed in yeast [10–14], however, an analysis of general trends in the gene expression response of human cells to extracellular cues and of their functional consequences on the regulation of human cell behavior has not been performed.
We undertook a functional analysis of global trends in the expression variance of human genes in response to extracellular cues. Expression variance of a gene can be defined as the frequency and magnitude of change in its mRNA levels in response to changing extracellular conditions and can be thought of as regulatability of a gene at the mRNA level. First, we report that human genes display a wide spectrum of EV under physiological conditions, with some genes showing very little variation in their mRNA levels, while some have extremely variable expression across a wide range of conditions. The EV pattern of genes strongly correlates with their promoter architecture, such that genes with lowest EV have open promoters with constitutive RNA polymerase occupancy, while those with highest EV have closed promoters with little or no RNA polymerase occupancy. Then, we show that this pattern of EV under physiological conditions reflects positioning of genes in the hierarchy of cell signaling, such that the most highly regulated genes are located at the apical parts of signaling hierarchy and are generally functionally more specialized. Finally, we discuss implications of these findings on our understanding of the generic mechanisms of regulation of cell behavior as it relates to restructuring of the intracellular protein interactome. This study uncovers some of the basic principles of transcriptional response in human cells and expands our understanding of conditional gene expression at the protein network level suggested by earlier studies on tissue-specific gene expression .
Calculating Expression Variance of human genes
It is possible that the EV values simply reflect basal tissue specific expression variations of genes and not the variability of their expression under different cellular conditions. In order to test this, we calculated tissue-specific EVs of genes using only samples in the ExpO dataset collected from breast, lung or colon, thereby obtaining EVs of genes for each individual tissue type. If the EVs reflect tissue-specific expression variations of genes, there should not be high correlation between tissue-specific EVs. However, there is a high correlation between breast and lung tissue-specific EVs (Additional File 1). Similarly high correlation was also observed between ovary and lung tissue EVs (Additional file 2, which indicates that EV mostly reflects variability of genes between different cellular and extracellular conditions rather than tissue-specific expression patterns of genes. We also tested correlation of EV values between different probes of the same gene, and find similar high correlation (Spearman's ρ = 0.45, n = 10,263, P <<10-16), indicating that the EV values identified here represent gene-specific variations of mRNA levels.
EV reflects gene regulation under varying extracellular conditions
To further confirm that our EV values reflect cellular response to varying extracellular conditions rather than being an artifact of tissue samples, we compiled an independent collection of microarray gene expression datasets from 14 different studies measuring responses of cultured human cells to various receptor ligands (EGF, heregulin, TGF-beta, TNF-alpha, interleukin 1, FGF2, arachidonic acid, thrombin, leukotriene, estradiol and sphingosine) (CK dataset, see Methods). We normalized each microarray sample in the dataset by their corresponding controls (i.e. no treatment conditions) and discarded the control samples, so that each sample in the CK dataset reflects fold-changes in response to the corresponding stimulus. Therefore, this dataset contains measurements of gene expression change in various human cell lines under 149 different stimulation conditions. The expression variance of genes calculated using the CK dataset is in a significantly high agreement with the EV values calculated using the ExpO dataset of 2158 tissue samples (Spearman's ρ = 0.69, see Additional file 3). Since EVCK values reflect fold changes of gene expression upon a large number of different stimulations, our observation indicates that EVexpo values reflect true expression variations of genes within cells under different extracellular environments, rather than an artifact of tissue- or cell type-specific expressional variations.
EV is not an artifact of mRNA abundance
Total mRNA expression levels of genes are extremely variable (spanning almost 4 logs), and this can substantially contribute to the variability of genes between different conditions. Indeed, EV of genes has a significant negative correlation with their average expression levels across the whole ExpO dataset (Spearman's ρ = -0.59 for EVexpo and -0.53 for EVCK), so that genes with low mRNA abundance are more likely to have variable expression. Therefore, it is possible that our observations above and below simply reflect the correlations of total expression levels of gene mRNAs rather than their variability. In order to test this, we calculated partial correlation between EVexpo and EVCK having controlled for average mRNA expression levels of genes and find that the correlation strength between these EV values calculated from different datasets is still significantly strong (partial Spearman's ρ = 0.58), indicating that the observed EV is not an artifact of mRNA abundance. In order to confirm this observation, we selected genes with similar average mRNA levels (300 < average expression < 350, n = 831), and tested if the correlation between EVexpo and EVCK is still high. Indeed, although the correlation of total mRNA levels with either EVCK or EVexpo is lost (Spearman's ρ values of 0.02 and -0.003, respectively), the correlation between EVCK and EVexpo is still significantly high (Spearman's ρ = 0.50, Additional file 4), which strongly suggests that expression variability is an intrinsic characteristic of genes rather than an artifact of their total mRNA abundance. Importantly, the correlations we present below are also reproducible having controlled for the total mRNA levels of genes (see below and Additional file 5).
EV reflects RNA Polymerase II promoter occupancy
Functional distinction of genes based on EV
We have previously shown organization of genes into separate modules based on their expression variation in yeast [10, 12]. We wanted to determine if expression variation in human genes has a functional significance similar to that observed in yeast. In order to answer this question, we constructed a comprehensive network of human genes based on their functional similarity, where each interaction is between two genes sharing a significant functional annotation from either Gene Ontology  or KEGG  (the Fun-Net, see Methods). Then, we tested whether subnetworks of genes with specific functional associations segregated based on their EV. In order to gain a comprehensive view of gene-gene association preferences in the Fun-Net based on their EV, we binned genes into 50 bins based on their EV and calculated interaction preferences between each bin pair in the Fun-Net. As expected, the heatmap of interaction preferences shows a clear clustering of low and high EV genes into separate functional categories (Figure 2B). This is not an artifact of the network connectivity, as this pattern is not observed in a network where node positions have been randomly shuffled (Additional file 7). Similarly analysis using different bin sizes did not significantly alter the outcome (Additional file 8). A similar high correlation and interaction preference pattern, albeit weaker, is observed when protein-protein interaction network is used for gene-gene interactions instead of the Fun-Net (Additional file 9). These observations show that human genes can be functionally separated based on their EV patterns. Low overall association of genes with low and high EV genes in either network suggests that the cellular functions performed by the low and high EV genes are distinct, similar to what we have shown for yeast.
Next, in order to see which cellular functions are represented by the high and low EV genes, we calculated relative enrichment of the top and bottom 500 genes within the EV distribution for specific GO functional categories. Figure 2C shows pie-charts of most enriched (hypergeometric distribution p-value < 10-5) functional categories in 500 genes with lowest or highest EV. Genes encoding for cellular functions pertaining to cellular homeostasis: mRNA transcription and processing, protein synthesis and proteasomal protein degradation, are the most significantly enriched functional categories among genes with lowest EV (Figure 2C). However, genes exhibiting highest EV are mainly composed of genes encoding proteins in the extracellular space, including extracellular matrix (ECM) components, growth factors and extracellular proteases. A similar pattern is identified using the EVCK values (Additional file 10), where the values reflect fold inductions of genes within the same cell line in response to a treatment. Therefore, the differential enrichment of high and low EV genes for, respectively, extracellular space and intracellular homeostasis genes reflects biological pattern of cellular response to extracellular conditions.
EV of signaling genes reflects their role in the signaling hierarchy
Interestingly, class II, which represents genes with PIC occupancy but no detectable transcription, is enriched for intracellular signaling genes in the RS level, not secreted factors, although this class has significantly high EV (see Figure 2A). This may indicate that class II contains condition-specific intracellular signaling genes, while classes I and III are enriched for constitutively expressed intracellular signaling genes. Indeed, genes with class II promoters contain high EV genes of the RS level, while classes I and III contain the low EV genes of the RS level (Figure 3D). Importantly, these observations suggest not only that genes coding for extra- and intra-cellular proteins can be distinguished based on their promoter architecture, but also that promoters of intracellular proteins among themselves are distinguished based on whether they are constitutive or condition-specific. In addition, while genes for condition-specific extracellular proteins are located within densely packed hypo-acetylated regions, condition-specific intracellular genes have relatively open promoters with a pre-assembled PIC. This suggests that regulation of transcription of genes coding for extracellular proteins may be fundamentally different from those coding for intracellular signaling genes (see Discussion).
EV reflects functional centrality
Discussion & Conclusions
Expressional variation of human genes
Computational studies in yeast combining large-scale gene expression data with protein interaction networks have revealed high level of modularity in the network with respect to transcriptional regulation [10, 12, 13, 19, 20]. However, with the exception of some recent studies [15, 21], such studies with human data have not been performed. Here, we report a study of global patterns in the expressional variation of human genes across a wide spectrum of conditions, and the functional significance of EV with respect to the regulation of signaling network architecture. Our findings were reproduced using two independent data compendiums, suggesting that these observations reflect true biological relationships. In addition, since variations in mRNA levels of genes have been shown to be in a relatively high agreement with corresponding variations in protein levels [22–24], the patterns of EV discovered in this study give insight into the patterns of regulation of signaling networks in response to extracellular stimuli.
Our results show that human genes are extremely variable in the extent of regulation of their mRNA levels. While some genes' mRNA levels are highly variable across many conditions, some show very tight expression patterns with very little variation. As expected, genes with lowest EV are those involved in cellular "housekeeping" functions, such as mRNA synthesis and processing as well as protein synthesis and degradation. In agreement with prior data about condition-specific genes [8, 25], genes with high EV mainly have "covered" promoters with reduced histone acetylation and no RNA polymerase pre-initiation complex (PIC) occupancy, while genes with low EV have high PIC occupancy and increased histone acetylation in their promoters.
Transcriptional regulation of intracellular and extracellular proteins
Our analyses correlating previous classification of genes into 4 distinct classes of promoters by Kim et al (2005)  revealed that there is a high concordance of EV values with their promoter architectures. Low EV genes are abundantly and actively transcribed, while high EV genes are generally not active. Most interestingly, high EV genes coding for intracellular signaling proteins have acetylated promoters with pre-assembled PIC, while high EV genes coding for extracellular proteins have hypo-acetylated promoters without PIC. Importantly, this may imply that the regulation of gene expression for extracellular proteins involves chromatin remodeling and PIC assembly, while that for intracellular proteins occurs at the level of RNA polymerase II elongation, rather than PIC assembly and chromatin remodeling. It has been reported that promoter-proximal pausing of the RNA polymerase II and its subsequent release for elongation is a major mechanism of regulation of human gene expression , which suggests that this mechanism may be employed for the regulation of intracellular proteins.
It should be noted that the study of Kim et al was performed on human fibroblasts, and therefore it could be argued that the classification of genes into distinct promoter classes may be specific to fibroblasts, despite the observed high correlation of the EV patterns with these classes. We find that EV of genes is highly similar between different tissues (ExpO dataset) and different conditions (CK dataset), suggesting that a common pool of condition-specific genes may exist, selective modulation of which may drive cell adaptation. Similarly, genes with lowest EV are primarily those with housekeeping functions in the cell, and are therefore likely to be expressed in all cell types. Therefore, it is likely that the overall chromatin architecture of most human promoters is also largely conserved between different cell types, and tissue and cell type-specific promoters may constitute a relative minority. This hypothesis is not far-fetched, as another recent study analyzing chromatin architecture around gene promoters in a number of different human cell lines reported more than 70% similarity in observed positioning of nucleosomes in promoters of different cell types .
Regulation of signaling network architecture
The observation that genes regulated most in response to extracellular stimuli are secreted factors and their receptors implies that regulation of cell behavior mostly involves modulation of the composition of the extracellular environment. Even the intracellular signaling proteins with high EV seem to be mainly those with specialized roles in the regulation of signaling and with fewer number of functional interactions. This indicates that the repertoire of the extracellular space and of their receptors mostly determines cell behavior, while the intracellular signaling hubs are mainly common for different cell types/conditions. Since in a scale-free network, such as the protein-protein interaction network , highly connected hubs play an important role in determining the overall architecture , our findings may suggest that the overall architecture of the signaling network is relatively stable across different conditions. Therefore, regulation of cell signaling during cell adaptation is mainly at the level of signaling inputs at the extracellular space, and minor highly specific rearrangements within the intracellular network. This in turn suggests that relatively same signaling network architecture allows for integration of various inputs to elicit a variety of cell fates, reminiscent of a multifunctional electronic circuit. A relatively stable network architecture where the hubs are involved in multiple processes may be evolutionarily more advantageous over a highly dynamic network architecture where hubs are condition-specific. It is interesting to note that similar conclusions have been drawn from recent studies on tissue-specific genes, where it was reported that tissue-specific proteins are enriched for extracellular proteins , and another reporting that tissue-specific proteins generally have less number of protein interactions . Therefore, it is possible that the regulatory principles in response to diverse external stimuli uncovered in this study also apply to tissue-specific modulation of cell behavior.
It can be argued that quantitating protein-protein interactions to show relative centrality of proteins may introduce artifacts of historically more studied proteins. However, we suggest it is a fair assumption that the distribution of well-studied proteins across the EV spectrum is relatively uniform so as to allow for the detection of statistically significant patterns.
The ExpO dataset was downloaded from the web site for Expression Project for Oncology (http://www.intgen.org/expo/). Each column in the final dataset of 2158 samples was first normalized by quantile normalization, and then each row was normalized by its median value and log2 transformed. EV values were determined as statistical variance value of a gene across all the samples in the normalized dataset. The CK compendium was derived from datasets in Gene Expression Omnibus: GDS649 (IL1 treatment of HUVEC cells), GDS1290 (TGF-beta treatment of Th1 and Th2 cells), GDS1249 (arachidonic acid treatment of dendritic cells), GDS2516 (interferon treatment of endothelial and fibroblast cells), GDS3215 (retinoic acid treatment of sebocyte cells), GDS1926 (leukotriene and thrombin treatment of endothelial cells), GDS2626 (EGF and HRG treatment of MCF7 cells), GDS2422 (FGF2 treatment of fibroblasts), GDS2484 (TNF-alpha treatment of endothelial cells), GDS2622 (EGF treatment of MCF10A cells), GDS3217 (estradiol treatment of MCF7), GDS2090 (sphingosine treatment of glioblastoma cell line), GDS855 (TGF-beta treatment of CD34+ cells) and GDS854 (TGF-beta treatment of a leukemia cell line). The columns in each dataset in the CK compendium were normalized by their respective control conditions (e.g. 0 time point), and columns for control conditions were discarded. Values were log2 transformed and each column was then normalized to have a mean of 0 and a variance of 1. EV values for each data compendium is given in Additional file 12.
Functional similarity interactions (Fun-Net) were constructed using Gene Ontology (GO) annotations as defined in the Entrez Gene database, and also metabolic pathway annotations in the KEGG database. Any two genes sharing a metabolic pathway annotation from KEGG were assigned an interaction. In the case of GO annotations, two genes were assigned an interaction if the overlap of their GO annotations was significant compared to the rest of the genes: s ij = |∩ G k |/ n, where s ij is the significance of overlap between genes i and j; G k is the set of genes that have the GO term k, where k belongs to the set of GO terms common to genes i and j, and n is the total number of genes. If s ij < 0.001, genes i and j were assigned an interaction. Protein-protein interactions were compiled from online databases HPRD , BIND , HomoMINT , Gene  and IntAct . For the signaling network, we compiled signaling interactions from KEGG, BioCarta (http://pid.nci.nih.gov/) and TRANSPATH , as well as through manual curation of some undirected protein-protein interactions. Transcription factor-target interactions were obtained from ORegAnno , TRANSFAC  and interactions in BIND classified as protein-DNA. Both networks are available from authors upon request.
Cell surface receptors and extracellular proteins were determined by combining genes with GO annotations as described in text. Receptors were assigned directly to GR. SF class was defined by determining extracellular proteins with direct signaling interactions with the GR group proteins. GM is defined as extracellular proteins with direct signaling interactions with the SF but not GR groups. RS are proteins with direct signaling interactions with GR and RS2 are those with direct signaling interactions with RS but not GR. Lists of genes within each hierarchy class is given in Additional file 13.
This work was supported in part by NIH grants U54 CA112970 and R01CA125109.
- van Steensel B: Mapping of genetic and epigenetic regulatory networks using microarrays. Nature genetics. 2005, 37 (Suppl): S18-24. 10.1038/ng1559View ArticlePubMedGoogle Scholar
- Bild AH, Potti A, Nevins JR: Linking oncogenic pathways with therapeutic opportunities. Nat Rev Cancer. 2006, 6 (9): 735-741. 10.1038/nrc1976View ArticlePubMedGoogle Scholar
- Slonim DK: From patterns to pathways: gene expression data analysis comes of age. Nature genetics. 2002, 32 (Suppl): 502-508. 10.1038/ng1033View ArticlePubMedGoogle Scholar
- Segal E, Friedman N, Kaminski N, Regev A, Koller D: From signatures to models: understanding cancer using microarrays. Nature genetics. 2005, 37 (Suppl): S38-45. 10.1038/ng1561View ArticlePubMedGoogle Scholar
- The ENCODE (ENCyclopedia Of DNA Elements) Project. Science. 2004, 306 (5696): 636-640. New York, NYGoogle Scholar
- Caron H, van Schaik B, van der Mee M, Baas F, Riggins G, van Sluis P, Hermus MC, van Asperen R, Boon K, Voute PA: The human transcriptome map: clustering of highly expressed genes in chromosomal domains. Science. 2001, 291 (5507): 1289-1292. New York, NYGoogle Scholar
- Birney E, Stamatoyannopoulos JA, Dutta A, Guigo R, Gingeras TR, Margulies EH, Weng Z, Snyder M, Dermitzakis ET, Thurman RE, et al.: Identification and analysis of functional elements in 1% of the human genome by the ENCODE pilot project. Nature. 2007, 447 (7146): 799-816. 10.1038/nature05874View ArticlePubMedGoogle Scholar
- Kim TH, Barrera LO, Zheng M, Qu C, Singer MA, Richmond TA, Wu Y, Green RD, Ren B: A high-resolution map of active promoters in the human genome. Nature. 2005, 436 (7052): 876-880. 10.1038/nature03877PubMed CentralView ArticlePubMedGoogle Scholar
- Carninci P, Kasukawa T, Katayama S, Gough J, Frith MC, Maeda N, Oyama R, Ravasi T, Lenhard B, Wells C: The transcriptional landscape of the mammalian genome. Science. 2005, 309 (5740): 1559-1563. New York, NYGoogle Scholar
- Komurov K, Gunes MH, White MA: Fine-scale dissection of functional protein network organization by statistical network analysis. PloS one. 2009, 4 (6): e6017- 10.1371/journal.pone.0006017PubMed CentralView ArticlePubMedGoogle Scholar
- Ihmels J, Levy R, Barkai N: Principles of transcriptional control in the metabolic network of Saccharomyces cerevisiae. Nature biotechnology. 2004, 22 (1): 86-92. 10.1038/nbt918View ArticlePubMedGoogle Scholar
- Komurov K, White M: Revealing static and dynamic modular architecture of the eukaryotic protein interaction network. Molecular systems biology. 2007, 3: 110- 10.1038/msb4100149PubMed CentralView ArticlePubMedGoogle Scholar
- Han JD, Bertin N, Hao T, Goldberg DS, Berriz GF, Zhang LV, Dupuy D, Walhout AJ, Cusick ME, Roth FP, et al.: Evidence for dynamically organized modularity in the yeast protein-protein interaction network. Nature. 2004, 430 (6995): 88-93. 10.1038/nature02555View ArticlePubMedGoogle Scholar
- Ge H, Liu Z, Church GM, Vidal M: Correlation between transcriptome and interactome mapping data from Saccharomyces cerevisiae. Nature genetics. 2001, 29 (4): 482-486. 10.1038/ng776View ArticlePubMedGoogle Scholar
- Bossi A, Lehner B: Tissue specificity and the human protein interaction network. Molecular systems biology. 2009, 5: 260- 10.1038/msb.2009.17PubMed CentralView ArticlePubMedGoogle Scholar
- Bolstad BM, Irizarry RA, Astrand M, Speed TP: A comparison of normalization methods for high density oligonucleotide array data based on variance and bias. Bioinformatics (Oxford, England). 2003, 19 (2): 185-193. 10.1093/bioinformatics/19.2.185View ArticleGoogle Scholar
- Ashburner M, Ball CA, Blake JA, Botstein D, Butler H, Cherry JM, Davis AP, Dolinski K, Dwight SS, Eppig JT, et al.: Gene ontology: tool for the unification of biology. The Gene Ontology Consortium. Nature genetics. 2000, 25 (1): 25-29. 10.1038/75556PubMed CentralView ArticlePubMedGoogle Scholar
- Kanehisa M, Goto S: KEGG: kyoto encyclopedia of genes and genomes. Nucleic acids research. 2000, 28 (1): 27-30. 10.1093/nar/28.1.27PubMed CentralView ArticlePubMedGoogle Scholar
- de Lichtenberg U, Jensen LJ, Brunak S, Bork P: Dynamic complex formation during the yeast cell cycle. Science. 2005, 307 (5710): 724-727. New York, NYGoogle Scholar
- Luscombe NM, Babu MM, Yu H, Snyder M, Teichmann SA, Gerstein M: Genomic analysis of regulatory network dynamics reveals large topological changes. Nature. 2004, 431 (7006): 308-312. 10.1038/nature02782View ArticlePubMedGoogle Scholar
- Cui Q, Yu Z, Purisima EO, Wang E: Principles of microRNA regulation of a human cellular signaling network. Molecular systems biology. 2006, 2: 46- 10.1038/msb4100089PubMed CentralView ArticlePubMedGoogle Scholar
- Varambally S, Yu J, Laxman B, Rhodes DR, Mehra R, Tomlins SA, Shah RB, Chandran U, Monzon FA, Becich MJ, et al.: Integrative genomic and proteomic analysis of prostate cancer reveals signatures of metastatic progression. Cancer cell. 2005, 8 (5): 393-406. 10.1016/j.ccr.2005.10.001View ArticlePubMedGoogle Scholar
- Newman JR, Ghaemmaghami S, Ihmels J, Breslow DK, Noble M, DeRisi JL, Weissman JS: Single-cell proteomic analysis of S. cerevisiae reveals the architecture of biological noise. Nature. 2006, 441 (7095): 840-846. 10.1038/nature04785View ArticlePubMedGoogle Scholar
- Weinstein JN: Integromic analysis of the NCI-60 cancer cell lines. Breast disease. 2004, 19: 11-22.PubMedGoogle Scholar
- Tirosh I, Barkai N: Two strategies for gene regulation by promoter nucleosomes. Genome research. 2008, 18 (7): 1084-1091. 10.1101/gr.076059.108PubMed CentralView ArticlePubMedGoogle Scholar
- Krumm A, Hickey LB, Groudine M: Promoter-proximal pausing of RNA polymerase II defines a general rate-limiting step after transcription initiation. Genes & development. 1995, 9 (5): 559-572.View ArticleGoogle Scholar
- Ozsolak F, Song JS, Liu XS, Fisher DE: High-throughput mapping of the chromatin structure of human promoters. Nature biotechnology. 2007, 25 (2): 244-248. 10.1038/nbt1279View ArticlePubMedGoogle Scholar
- Barabasi AL, Oltvai ZN: Network biology: understanding the cell's functional organization. Nature reviews. 2004, 5 (2): 101-113. 10.1038/nrg1272View ArticlePubMedGoogle Scholar
- Albert R, Jeong H, Barabasi AL: Error and attack tolerance of complex networks. Nature. 2000, 406 (6794): 378-382. 10.1038/35019019View ArticlePubMedGoogle Scholar
- Winter EE, Goodstadt L, Ponting CP: Elevated rates of protein secretion, evolution, and disease among tissue-specific genes. Genome research. 2004, 14 (1): 54-61. 10.1101/gr.1924004PubMed CentralView ArticlePubMedGoogle 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, D411-414. 34 DatabaseGoogle Scholar
- Bader GD, Donaldson I, Wolting C, Ouellette BF, Pawson T, Hogue CW: BIND--The Biomolecular Interaction Network Database. Nucleic acids research. 2001, 29 (1): 242-245. 10.1093/nar/29.1.242PubMed 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 research. 2007, D572-574. 35 DatabaseGoogle Scholar
- Maglott D, Ostell J, Pruitt KD, Tatusova T: Entrez Gene: gene-centered information at NCBI. Nucleic acids research. 2007, D26-31. 35 DatabaseGoogle 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, D561-565. 35 DatabaseGoogle Scholar
- Choi C, Krull M, Kel A, Kel-Margoulis O, Pistor S, Potapov A, Voss N, Wingender E: TRANSPATH-A High Quality Database Focused on Signal Transduction. Comparative and functional genomics. 2004, 5 (2): 163-168. 10.1002/cfg.386PubMed CentralView ArticlePubMedGoogle Scholar
- Griffith OL, Montgomery SB, Bernier B, Chu B, Kasaian K, Aerts S, Mahony S, Sleumer MC, Bilenky M, Haeussler M: ORegAnno: an open-access community-driven resource for regulatory annotation. Nucleic acids research. 2008, D107-113. 36 DatabaseGoogle Scholar
- Wingender E, Chen X, Hehl R, Karas H, Liebich I, Matys V, Meinhardt T, Pruss M, Reuter I, Schacherer F: TRANSFAC: an integrated system for gene expression regulation. Nucleic acids research. 2000, 28 (1): 316-319. 10.1093/nar/28.1.316PubMed 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.