Network and matrix analysis of the respiratory disease interactome
© Garcia et al.; licensee BioMed Central Ltd. 2014
Received: 25 November 2013
Accepted: 10 March 2014
Published: 22 March 2014
Although respiratory diseases exhibit in a wide array of clinical manifestations, certain respiratory diseases may share related genetic mechanisms or may be influenced by similar chemical stimuli. Here we explore and infer relationships among genes, diseases, and chemicals using network and matrix based clustering methods.
In order to better understand and elucidate these shared genetic mechanisms and chemical relationships we analyzed a comprehensive collection of gene, disease, and chemical relationships pertinent to respiratory disease, using network and matrix based analysis approaches. Our methods enabled us to analyze relationships and make biological inferences among over 200 different respiratory and related diseases, involving thousands of gene-chemical-disease relationships.
The resulting networks provided insight into shared mechanisms of respiratory disease and in some cases suggest novel targets or repurposed drug strategies.
KeywordsInteractome Networks Respiratory diseases Lung disease
The capability to catalog interactions among diseases, chemicals, and genes into well-curated databases offers a collective knowledge of experimental results that has great potential for the generation of hypotheses and meta-analyses. To date, many biological databases have been established to catalog relationships among genes, diseases, and chemicals. Many of these databases focus on one particular type of relational interaction, ranging from protein-protein interaction databases, gene-chemical databases, and disease-gene databases, and are often constructed using data mining methods complemented by manual curation. The described databases, in many instances, serve as the foundation for a wide array of predictive and analytical methods to examine interactions. They can also be extended to analyze interactions among overarching themes, including analyzing gene-chemical interactions within the context of a given set of diseases or protein-protein interactions within the context of peptide recognition[5, 6]. Integration of multiple sources and types of relational data remains an important and challenging research area with great potential toward the development of furthering our understanding complex diseases and interactions.
Each year over 400,000 deaths occur in the United States as a result of respiratory and related diseases (RRD). Given the high prevalence and importance of lung and respiratory diseases, we hypothesized that a better understanding of the respiratory gene-chemical-disease interactome would lead to better understanding of the molecular mechanisms of lung disease, including the environmental and drug influences, and more importantly, may lead to new treatment or intervention strategies. In this study, we focus our efforts on the analysis of gene-disease-chemical relationships, in order to elucidate and infer novel interactions and to understand biology pertinent to respiratory diseases using network and matrix-based methods.
Current network and matrix-based analyses of disease relationships has relied heavily on gene or protein-centric examinations[8–11], neglecting chemical features that may also influence disease. Likewise, network analysis techniques have often been developed and utilized to examine gene or protein relationships among diseases, but often neglect environmental or chemical factors that may influence disease. In cases where genes, diseases, and chemicals have been analyzed, often the networks are decoupled to allow for the analysis of a single entity or relationship type, such as the effect of a drug on a gene network or the elucidation of molecular mechanisms in disease[13–15]. Host-pathogen studies have also largely focused on a single relational type, predominantly protein-protein interaction relationships. Here we apply methods to investigate gene-chemical-disease networks, in order to better understand the genetic and chemical contributors of diseases, elucidating novel biology and helping to further understand shared disease pathology.
Results and discussion
In order to compile a comprehensive dataset to examine gene, disease, and chemical relationships pertinent to respiratory disease, we extracted information from the Comparative Toxicogenomics Database (CTD) and the Human Protein Reference Database (HPRD). CTD houses manually curated information pertinent to gene-disease-chemical relationships for a wide variety of diseases, and HPRD houses information focusing on protein-protein interactions from a wide array of experiments in humans and other model organisms. CTD offers a conservative and expert curated source of interactions to form networks, and HPRD uses the same normalized gene names as CTD.
We compiled and filtered our in-house database in two ways. The first database, we refer to as the whole respiratory network (Additional file1: Table S1), and the second database we refer to as the therapeutic network (Additional file2: Table S2). The whole respiratory network represents disease-gene, disease-chemical, chemical-gene, and gene-gene interactions associated with respiratory diseases. The therapeutic network, in contrast, consists of a subset of the respiratory network, containing only chemicals with curated therapeutic interactions with diseases and the genes that interact with those chemicals. These curated therapeutic interactions are established using the “DirectEvidence” field from CTD. This network was called the therapeutic network as a reference to this inclusion criterion. In addition to the therapeutic inclusion criteria, chemical-chemical interactions were also included based upon curated chemical relationships derived from chemical gene-interaction information. Gene-gene interactions were established using the HPRD database.
To assess the directionality of chemical-gene interactions, the uniqueness of chemical-gene and gene-chemical interactions were assessed. First, chemicals with disease interactions were batched queried using CTD, with an output of curated chemical-gene interactions. Second, genes with disease interactions were batched queried using CTD, with an output of curated gene-chemical interactions. The intersection between these two sets was then calculated. In the whole respiratory network, there were 27075 total chemical-gene and gene-chemical linkages with 13543 remaining after accounting for bi-directionality of interactions. Given the small percentage of directional linkages (~0.05%), all links were treated as bi-directional.
The type of interaction was established for disease-chemical, disease-gene, and chemical-gene interactions. For disease-chemical and disease-gene interactions, there were three types of interactions based upon CTD curation: therapeutic, marker/mechanism, and both therapeutic and marker/mechanism. Chemical-gene interactions had three major effects and one minor effect based upon CTD curation. The major effects are increasing, decreasing, and affecting expression or activity. The minor effect is based upon the type of protein modification imparted by the chemical onto the protein. The list of protein modification includes: ubiquitination, phosphorylation, oxidation, cleavage, methylation, hydrolysis, hydroxylation, glycosylation, glucuronidation, acetylation, nitrosation, ribosylation.
To establish chemical-chemical linkages and the type of gene-chemical linkage, CTD was used. Chemical-gene interactions were extracted with a query specifying interaction type. Co-interactions between multiple chemicals and a gene were extracted from this list and chemical-chemical linkages were established if two chemicals had a curated co-occurrence with a gene. A co-occurrence was determined when a secondary chemical appeared in the interaction characteristics between a chemical and a gene. The type of linkage between the two chemicals was classified using the same type of link used to classify chemical-gene interaction in which the co-occurrence appeared. As there is often discordance between the naming of chemicals, especially those with pharmaceutical implications, a chemical reaction database and drug interaction database were not utilized for establishing chemical-chemical interactions.
Evaluation of protein-protein interaction network clustering methods is generally performed through the comparison of gold standard regulatory networks or pathways. Since an analogous gold standard gene-chemical-disease network does not exist, for us to evaluate clustering methods, we selected high performing methods used for clustering protein-protein interaction networks, with the added stipulation that their output is scalable to a more sparse and dissimilar network. MCODE and MCL, two widely accepted and utilized clustering methods were tested for scalability when adding additional node types[11, 17–19]. The gene-gene portion of the network was used as a baseline for the types and sizes of sub-networks that can be expected in an ideal situation. In the gene-gene network, both algorithms performed similarly with median sub-network sizes of 4 for MCODE and 3 for MCL. In the larger sub-networks both methods displayed highly interconnected clusters. In the therapeutic network, however, the clustering methods performed much differently. MCODE had a median sub-network size of 18, while still maintaining the highly interconnected networks, and MCL had a median cluster size of 3, and no longer exhibited an interconnected feature. We also applied hierarchical clustering, utilizing a Pearson’s correlation coefficient. Pearson’s has been shown to be a highly robust unsupervised correlation that performs well under a multitude of protein-protein interaction analyses, from identifying regulatory networks to identifying groups of proteins with shared functions[20, 21]. A lack of a gold standard gene-disease-chemical network is also why no semi-supervised or supervised methods were chosen.
For the whole respiratory network, nodes were input based upon type (disease, chemical, gene) and edges based upon types of nodes involved (disease-gene, gene-gene, chemical-gene, disease-chemical) into Cytoscape, creating a network of 1,830 nodes and 17,275 edges. This network became a test-bed for methods to improve subsequent analyses including constructing networks with only one type of edge, and networks with filtered diseases, chemicals, and genes of interest. These tests led to the creation of both a gene-disease sub-network and the therapeutic chemical network.
Network nodes and links
Inferred interaction summary
One of these sub-networks, shown in Figure 4A, contains non-small-cell lung carcinoma and closely interacting genes and chemicals. From this sub-network, three links were analyzed in greater detail: aspirin - EGFR, acetaminophen - non-small-cell carcinoma, and piroxicam - non-small-cell carcinoma. Aspirin - EGFR is an inferred link in this sub-network that was added as a direct link to an update of CTD that occurred after the creation of this network. There was strong support in literature for aspirin promoting EGFR inhibitors, enough for a curated interaction between these two elements[26, 27]. This link represents a verified prediction both by literature and by CTD, representing the effectiveness of using sub-networks to find novel links. Acetaminophen - Non-small-cell lung carcinoma is a link that has negative support in literature. In studies involving testing multiple anti-inflammatory drugs for change in non-small-cell lung carcinoma outcome, they found no correlation between Acetaminophen and change in prognosis. The negative support for this link shows that while sub-networks offer a starting point for testing inferred interactions, not all of the nodes will have a direct link. Lastly, Piroxicam - non-small-cell carcinoma had direct and indirect literature support for this link[29, 30]. There was increased immune function in lung cancer patients that had piroxicam added to their drug regimens. Also, piroxicam showed decreased tumorigenesis in mice with colon cancer, suggesting this link might be present in other cancers as well. This link represents a possible avenue for further research. There is evidence to support that there are beneficial effects of piroxicam on non-small-cell carcinoma prognosis; however, the full effects of this interaction are not well understood.
Analysis of sub-networks also presents the ability to find links for similar or comorbid diseases. In the cystic fibrosis sub-network, Figure 4B, dobutamine interacts with both of cystic fibrosis’ genes in the sub-network, suggesting a link between dobutamine and cystic fibrosis. Upon searching the literature, dobutamine, especially in combination with nitric oxide, improves pulmonary hypertension in cystic fibrosis patients, a common comorbidity. CTD neither contains a link between dobutamine and cystic fibrosis nor dobutamine and pulmonary hypertension.
Jaccard similarity assessment
A more systematic evaluation of the relationship between Jaccard similarity and identifying novel links was performed on a human signaling network. Protein-Protein interactions from the human signaling network were selected based upon both the interacting genes being present in the therapeutic network while their interaction was not present in the network. While self-interacting genes were utilized in generating Jaccard similarity values, they were excluded from both the background and the human signaling network during the analysis. This is due to the fact there is no way to distinguish between likely self-interactions and unlikely self-interactions using a similarity measure that will always be 1.0 in the case of a self-interaction. This selection resulted in 1057 additional interactions for use in validation.
The assertion that subclusters can identify nodes with similar properties can be used for predicting interactions by analyzing overlap between cluster nodes and their shared interactions. In a cluster containing SEPP1, GJB1, SELENBP1, SLC22A18, A2M, and PDFGA, five out of the six genes in this cluster have an association with lung neoplasms. PDGFA, the gene not linked with lung neoplasms, has associations with breast, prostate, head and neck, and pancreas cancers. In addition, PDGFA increases with asbestos exposure, a chemical linked to mesothelioma. This increase is also associated with tumorigenicity, supporting the assertion that PDFGA is also a marker for lung neoplasms.
Ebastine, levocabastine, hydroxyzine, SUN1334H, azelastine, olopatadine, cetirizine, desloratadine, sho-seiryu-to, epinastine, and tripolidine are a group of anti-histamine drugs that target HRH1, all of which also have interactions with rhinitis. These anti-histamine drugs also have anti-inflammatory properties, revealed by seven drugs having links to IL4, four having links to IL5, and four having links to IL8. This is supported by a study that shows various anti-histamines having anti-inflammatory properties in rhinitis pathology.
MT2, MT1, CCL9, CCL8, ECM1, and SLC39A4 represent a diverse cluster of two metallothionein proteins, two macrophage proteins, one extracellular matrix protein, and one zinc transporter protein. Many of these genes regulate metal concentrations within cells and are linked to respiratory hypersensitivity. Out of the five shared chemicals, only acetaminophen is linked to respiratory hypersensitivity. However, four out of these five chemicals have links to asthma, suggesting they may play a greater role in respiratory hypersensitivity in general. This hypothesis is supported by the fact that zinc deficiency alters respiratory epithelium in allergic response of mice.
Node-edge versus matrix
While these two approaches take the same input, clustering produces two distinct results. Only eight of the eighteen sub-networks contained a cluster from the matrix where at least 50% of the nodes present in the matrix cluster were also present in the sub-network. Most of the matrix clusters that overlapped with the sub-networks contained only two or three nodes. However, one sub-network contained 11 of the 28 nodes in one matrix subcluster, making it the most nodes shared between a sub-network and a matrix cluster. These differences can be attributed to both network construction and the types of interactions that are obtained from each approach. Given the sparsity of the network, especially in chemical-chemical interactions, and the lack of disease-disease interactions, clustering coefficients and pairwise comparisons produce non-overlapping results. Clustering coefficients from node-edge based approaches represent closely interacting genes, chemicals, and diseases. These closely interacting nodes offer avenues of exploration for finding novel interactions. Pairwise comparisons from matrixes represent nodes that share the same interaction profile. This interaction profile can then be used for determining both biological meaning and novel interactions for any pairs between the cluster nodes and the interaction profile nodes. Thus, these two approaches offer a complimentary analysis strategy for sparse networks, enabling elucidation of both novel interactions and increasing our biological understanding of node clusters.
The second distinction these two approaches offer is in the visualization of interactions. Node-edge network approaches illustrate which nodes form a sub-network, which nodes interact within these sub-networks, and the types of interactions between each node, giving an all encompassing view of the sub-network. Matrix-based approaches provide a broader view of interactions, offering a tool for visualizing not only how similar nodes and clusters are to each other, but also the interactions nodes share outside of their individual clusters.
Current network analyses of disease are still highly focused on gene and protein-based networks, neglecting environmental and drug effects that contribute to the pathophysiology of a disease or sets of diseases. Our proposed methods integrate both the chemical and disease entities into network and matrix-based analyses, allowing for a more complete systems understanding of the underlying biology. With this addition of multiple different entity types comes the lack of a gold standard for identifying specific genes, chemicals, and diseases that should cluster together, providing a similar role as the curated regulatory and pathway networks used to establish accuracy in protein-protein and gene-gene network analyses.
In order to better investigate complex and sparse networks, such as the respiratory disease interactome, a multi-method approach utilizing methods proven effective in gene-gene and protein-protein network-based analyses has proven useful to elucidate and investigate different network properties and the underlying biological context. In this case we have used two approaches: a node-edge-based clustering coefficient with Jaccard similarity comparison approach applied to traditional networks, and a matrix-based Pearson’s correlation coefficient with hierarchical clustering approach. This allows identification of closely interacting diseases, chemicals, and genes, as well as similar interaction profiles either within or between these same elements of interest. These two approaches help facilitate investigations on the underlying biology for a given disease, pathophysiology similarities across diseases, and chemicals that may have a therapeutic indication outside of their original use.
The shared interactome of four therapeutic antibiotics (ofloxacin, amoxicillin clavulanate, clarithromycin, and azizthromycin, (Figure 9) allows for an inference of interaction between azithromycin and tuberculosis based upon the interaction profile of a cluster generated by hierarchically clustering a Pearson’s correlation coefficient matrix. This profile represents the layering of diseases, chemicals, genes and the interactions between them, showing that while azithromycin has no known anti-Mycobacterium tuberculosis properties, it does have pro-host immune properties that may have therapeutic merit for tuberculosis treatment.
These methods are also useful for finding drug targets. The shared interactome of pulmonary fibrosis and asthma (Figure 1) demonstrates that Th2 cell inflammation is an important factor in both of these diseases, where a drug that improves the outcomes in one of these diseases may also be useful for the other disease. Looking at these interactomes provides a broader context for drug discovery and drug repurposing.
Chemical, gene, and disease interactomes offer a novel approach to not only identify shared biology among diseases, but also offer a method for identifying possible new drug targets and repurposed drug strategies. Layering additional interaction information, additional databases, and additional analysis techniques will allow for a more complete systems-based analysis that will extend to any complex disease interactome.
Respiratory diseases and the curated chemical and genes interactions with these diseases were extracted from CTD using the January 9, 2012 database version. Curated chemical-gene interactions were extracted from batch queries using the chemicals and genes associated with respiratory diseases. Genes, chemicals, and their associated links that did not contain a link to a respiratory disease were removed from the list. Duplicates of gene-chemical and chemical-gene links were also removed from this list. Gene-gene interactions were established using the April 13, 2010 version of the HPRD database. Genes and their associated interactions were excluded from the list if they did not contain a direct link to a respiratory disease. These interactions were further specialized by including only chemicals with therapeutic interactions to respiratory diseases in a therapeutic network, with the “therapeutic” name stemming from including only chemicals with at least one therapeutic indication. The therapeutic indication for a chemical is determined from the by the direct evidence field from CTD. Genes were then excluded if they did not contain a link to one of these therapeutic chemicals. Chemical-chemical links and chemical-gene interaction characteristics for the therapeutic network were established using the February 10, 2012 version of CTD. Chemical-chemical links were established using co-occurrence of chemicals in chemical-gene interactions. A chemical was established as co-occurring when a secondary chemical appeared in the interaction characteristics of chemical-gene interactions. A triple was stored for each interaction, including both interacting nodes and the type of interaction between them.
Network and matrix visualization
A file containing the triples of interactions and a file containing the type of node (chemical, gene, disease) were loaded into Cytoscape. Nodes were colored based upon their type, with chemicals represented as blue, genes as black, and diseases and orange. Interactions were colored based upon interaction characteristics, with positive interactions as green, negative interactions as red, mixed interactions as purple, and additional characteristics as increasing intensity.
A binary interaction matrix between nodes was created using the network construction file containing interaction triples. A value of 1 was used for any interaction type between nodes and a value of 0 was used for a lack of interaction between nodes. This binary interaction matrix was visualized by creating a bitmap of clustered interactions and the resulting dendrograms by using TreeView.
Network and matrix clusters
MCODE, a Cytoscape plugin, was used to generate each of the sub-networks[17, 22]. A degree cutoff of 2, and node score cutoff of 0.2, a k-core of 2, and a max depth of 100 were used as the MCODE parameters for generating clusters.
Cluster 3.0 was used to generate clusters for this matrix. An uncentered similarity with average linkage was used to calculate the hierarchical clustering. Similarity scores of 0.4 and 0.7 were used for creating clusters, based upon inflection points Figure 7.
Jaccard similarity coefficients were generated for both the therapeutic network and for sub-networks using the following formula:. This formula calculates the intersection of the two sets divided by their union. A set, in all cases, is all the nodes that interact with a given node, including any self-interactions. The intersection of two nodes is all shared interactions between the two nodes, with the union of the two nodes being all the nodes that interact with at least one of the nodes of interest. For the entire therapeutic network, a Mann–Whitney U test was run with the alternative hypothesis that linked nodes are more similar than unlinked nodes. For sub-networks, ranks of Jaccard coefficients were calculated using the individual sub-network that a node pair come from and then compared to the evidence of there being an interaction.
Sub-networks were used to assess the stability of the network in respect to changes in Jaccard coefficient. For a given sub-network, an additional network was generated for each missing edge. In each of these networks one additional edge was added between two existing unlinked nodes, creating a unique set of networks. Jaccard coefficients were then generated for each additional network. Two-sample Kolmogorov-Smirnov tests were used to assess whether or not the distribution of the original sub-network and the altered sub-networks was shifted. This was done for each of the sub-networks and their corresponding altered networks. The null hypothesis was that the Jaccard coefficient distribution of the network with an additional edge is the same as the unaltered sub-network, with the alternative hypothesis being that the distribution is shifted.
Original network parsing to establish interactions between nodes was done using perl version 5.12.4 on Mac OSX 10.7. This includes parsing interactions between genes, chemicals, and diseases, finding which chemicals have co-interactions with genes, finding unique interactions and directional interactions between chemicals and genes, finding interaction characteristics for disease-gene and disease-chemical interactions, and selecting inclusion criteria for interactions of interest to develop each network.
Further network parsing, matrix construction, and dendrogram parsing was done using C#/.NET 4.0 on a Windows 7 machine. This includes finding specific interaction characteristics for chemical-gene and chemical-chemical interactions, construction of the interaction matrix, visualization of the interaction matrix, and extracting clusters based upon a threshold from the output from Cluster 3.0.
BG acknowledges support from a NLM Institutional Training Grant, NIH 5T15LM009451, and from NICTA, which is funded by the Australian Government through the Department of Communications and the Australian Research Council through the ICT Centre of Excellence Program. All authors acknowledge support from a Translational Research Inititative Grant from National Jewish Health, MS acknowledges support from the Boettcher Foundation Webb-Waring Award, and BG thanks Cheng Soon Ong and Melissa Davis of NICTA in Melbourne and Sonia Leach of National Jewish Health for our discussions. BG also thanks Joseph Cursons of NICTA in Melbourne for his help with matrix visualization.
- Keshava Prasad TS, Goel R, Kandasamy K, Keerthikumar S, Kumar S, Mathivanan S, Telikicherla D, Raju R, Shafreen B, Venugopal A, Balakrishnan L, Marimuthu A, Banerjee S, Somanathan DS, Sebastian A, Rani S, Ray S, Harrys Kishore CJ, Kanth S, Ahmed M, Kashyap MK, Mohmood R, Ramachandra YL, Krishna V, Rahiman BA, Mohan S, Ranganathan P, Ramabadran S, Chaerkady R, Pandey A: Human protein reference database–2009 update. Nucleic Acids Res. 2009, 37: D767-D772. 10.1093/nar/gkn892.PubMed CentralView ArticlePubMedGoogle Scholar
- Baxevanis AD: Searching Online Mendelian Inheritance in Man (OMIM) for information for genetic loci involved in human disease. Current protocols in bioinformatics. 2003, 35 (9.13): 1--9. 13.15Google Scholar
- Goto S, Okuno Y, Hattori M, Nishioka T, Kanehisa M: LIGAND: database of chemical compounds and reactions in biological pathways. Nucleic Acids Res. 2002, 30: 402-404. 10.1093/nar/30.1.402.PubMed CentralView ArticlePubMedGoogle Scholar
- Davis AP, King BL, Mockus S, Murphy CG, Saraceni-Richards C, Rosenstein M, Wiegers T, Mattingly CJ: The comparative toxicogenomics database: update 2011. Nucleic Acids Res. 2011, 39: D1067-D1072. 10.1093/nar/gkq813.PubMed CentralView ArticlePubMedGoogle Scholar
- Lamb J, Crawford ED, Peck D, Modell JW, Blat IC, Wrobel MJ, Lerner J, Brunet JP, Subramanian A, Ross KN, Reich M, Hieronymus H, Wei G, Armstrong SA, Haggarty SJ, Clemons PA, Wei R, Carr SA, Lander ES, Golub TR: The connectivity Map: using gene-expression signatures to connect small molecules, genes, and disease. Science. 2006, 313: 1929-1935. 10.1126/science.1132939.View ArticlePubMedGoogle Scholar
- Tong AH, Drees B, Nardelli G, Bader GD, Brannetti B, Castagnoli L, Evangelista M, Ferracuti S, Nelson B, Paoluzi S, Quondam M, Zucconi A, Hogue CW, Fields S, Boone C, Cesareni G: A combined experimental and computational strategy to define protein interaction networks for peptide recognition modules. Science. 2002, 295: 321-324. 10.1126/science.1064987.View ArticlePubMedGoogle Scholar
- American Lung Association: State of lung disease in diverse communities. 2010Google Scholar
- Kaneko Y, Yatagai Y, Yamada H, Iijima H, Masuko H, Sakamoto T, Hizawa N: The search for common pathways underlying asthma and COPD. Int J Chron Obstruct Pulmon Dis. 2013, 8: 65-78.PubMed CentralPubMedGoogle Scholar
- Dockstader K, Nunley K, Karimpour-Fard A, Medway A, Nelson P, Port JD, Liggett SB, Bristow MR, Sucharov CC: Temporal analysis of mRNA and miRNA expression in transgenic mice overexpressing Arg- and Gly389 polymorphic variants of the beta1-adrenergic receptor. Physiol Genomics. 2011, 43: 1294-1306. 10.1152/physiolgenomics.00067.2011.PubMed CentralView ArticlePubMedGoogle Scholar
- Janjic V, Przulj N: Biological function through network topology: a survey of the human diseasome. Brief Funct Genomics. 2012, 11: 522-532. 10.1093/bfgp/els037.View ArticlePubMedGoogle Scholar
- Islam MF, Hoque MM, Banik RS, Roy S, Sumi SS, Hassan FM, Tomal MT, Ullah A, Rahman KM: Comparative analysis of differential network modularity in tissue specific normal and cancer protein interaction networks. J Clin Bioinformatics. 2013, 3: 19-10.1186/2043-9113-3-19.View ArticleGoogle Scholar
- Barzel B, Barabasi AL: Network link prediction by global silencing of indirect correlations. Nat Biotechnol. 2013, 31: 720-725. 10.1038/nbt.2601.PubMed CentralView ArticlePubMedGoogle Scholar
- Iorio F, Saez-Rodriguez J, Bernardo D: Network based elucidation of drug response: from modulators to targets. BMC Syst Biol. 2013, 7: 139-10.1186/1752-0509-7-139.PubMed CentralView ArticlePubMedGoogle Scholar
- Wang X, Wei X, Thijssen B, Das J, Lipkin SM, Yu H: Three-dimensional reconstruction of protein networks provides insight into human genetic disease. Nat Biotechnol. 2012, 30: 159-164. 10.1038/nbt.2106.PubMed CentralView ArticlePubMedGoogle Scholar
- Yeh SH, Yeh HY, Soo VW: A network flow approach to predict drug targets from microarray data, disease genes and interactome network - case study on prostate cancer. J Clin Bioinformatics. 2012, 2: 1-View ArticleGoogle Scholar
- Gulbahce N, Yan H, Dricot A, Padi M, Byrdsong D, Franchi R, Lee DS, Rozenblatt-Rosen O, Mar JC, Calderwood MA, Baldwin A, Zhao B, Santhanam B, Braun P, Simonis N, Huh KW, Hellner K, Grace M, Chen A, Rubio R, Marto JA, Christakis NA, Kieff E, Roth FP, Roecklein-Canfield J, Decaprio JA, Cusick ME, Quackenbush J, Hill DE, Münger K, Vidal M, Barabási AL: Viral perturbations of host networks reflect disease etiology. PLoS Comput Biol. 2012, 8: e1002531-10.1371/journal.pcbi.1002531.PubMed CentralView ArticlePubMedGoogle Scholar
- Bader GD, Hogue CW: An automated method for finding molecular complexes in large protein interaction networks. BMC Bioinforma. 2003, 4: 2-10.1186/1471-2105-4-2.View ArticleGoogle Scholar
- Enright AJ, Van Dongen S, Ouzounis CA: An efficient algorithm for large-scale detection of protein families. Nucleic Acids Res. 2002, 30: 1575-1584. 10.1093/nar/30.7.1575.PubMed CentralView ArticlePubMedGoogle Scholar
- Nguyen P, Srihari S, Leong H: Identifying conserved protein complexes between species by constructing interolog networks. BMC Bioinforma. 2013, 14: S8-View ArticleGoogle Scholar
- Maetschke SR, Madhamshettiwar PB, Davis MJ, Ragan MA: Supervised, semi-supervised and unsupervised inference of gene regulatory networks. Brief Bioinform. 2013, 15: 195-211.PubMed CentralView ArticlePubMedGoogle Scholar
- Trinidad JC, Thalhammer A, Burlingame AL, Schoepfer R: Activity-dependent protein dynamics define interconnected cores of co-regulated postsynaptic proteins. Mol Cell Proteomics: MCP. 2013, 12: 29-41. 10.1074/mcp.M112.019976.PubMed CentralView ArticlePubMedGoogle Scholar
- Smoot ME, Ono K, Ruscheinski J, Wang PL, Ideker T: Cytoscape 2.8: new features for data integration and network visualization. Bioinformatics. 2011, 27: 431-432. 10.1093/bioinformatics/btq675.PubMed CentralView ArticlePubMedGoogle Scholar
- Lee CG, Cho SJ, Kang MJ, Chapoval SP, Lee PJ, Noble PW, Yehualaeshet T, Lu B, Flavell RA, Milbrandt J, Homer RJ, Elias JA: Early growth response gene 1-mediated apoptosis is essential for transforming growth factor beta1-induced pulmonary fibrosis. J Exp Med. 2004, 200: 377-389. 10.1084/jem.20040104.PubMed CentralView ArticlePubMedGoogle Scholar
- Wills-Karp M: Interleukin-13 in asthma pathogenesis. Immunol Rev. 2004, 202: 175-190. 10.1111/j.0105-2896.2004.00215.x.View ArticlePubMedGoogle Scholar
- Madhamshettiwar PB, Maetschke SR, Davis MJ, Reverter A, Ragan MA: Gene regulatory network inference: evaluation and application to ovarian cancer allows the prioritization of drug targets. Genome Med. 2012, 4: 41-10.1186/gm340.PubMed CentralView ArticlePubMedGoogle Scholar
- Becker JC, Muller-Tidow C, Stolte M, Fujimori T, Tidow N, Ilea AM, Brandts C, Tickenbrock L, Serve H, Berdel WE, Domschke W, Pohle T: Acetylsalicylic acid enhances antiproliferative effects of the EGFR inhibitor gefitinib in the absence of activating mutations in gastric cancer. Int J Oncol. 2006, 29: 615-623.PubMedGoogle Scholar
- Selvendiran K, Bratasz A, Tong L, Ignarro LJ, Kuppusamy P: NCX-4016, a nitro-derivative of aspirin, inhibits EGFR and STAT3 signaling and modulates Bcl-2 proteins in cisplatin-resistant human ovarian cancer cells and xenografts. Cell Cycle. 2008, 7: 81-88. 10.4161/cc.7.1.5103.PubMed CentralView ArticlePubMedGoogle Scholar
- Van Dyke AL, Cote ML, Prysak G, Claeys GB, Wenzlaff AS, Schwartz AG: Regular adult aspirin use decreases the risk of non-small cell lung cancer among women. Canc Epidemiol Biomarkers Prev: a Pub of the Am Assoc for Cancer Res, cosponsored by the Am Soc of Preventive Oncolgy. 2008, 17: 148-157.View ArticleGoogle Scholar
- Braun DP, Bonomi PD, Taylor SG, Harris JE: Modification of the effects of cytotoxic chemotherapy on the immune responses of cancer patients with a nonsteroidal, antiinflammatory drug, piroxicam. A pilot study of the Eastern Cooperative Oncology Group. J Biol Response Modif. 1987, 6: 331-345.Google Scholar
- Palmerini E, Fan K, Yang K, Risio M, Edelmann W, Lipkin M, Biasco G: Piroxicam increases colon tumorigenesis and promotes apoptosis in Mlh1 +/− /Apc1638(N/+) mice. Anticancer Res. 2007, 27: 3807-3812.PubMedGoogle Scholar
- Vizza CD, Rocca GD, Roma AD, Iacoboni C, Pierconti F, Venuta F, Rendina E, Schmid G, Pietropaoli P, Fedele F: Acute hemodynamic effects of inhaled nitric oxide, dobutamine and a combination of the two in patients with mild to moderate secondary pulmonary hypertension. Critical Care. 2001, 5: 355-361. 10.1186/cc1069.PubMed CentralView ArticlePubMedGoogle Scholar
- Zaman N, Li L, Jaramillo ML, Sun Z, Tibiche C, Banville M, Collins C, Trifiro M, Paliouras M, Nantel A, O'Connor-McCourt M, Wang E: Signaling network assessment of mutations and copy number variations predict breast cancer subtype-specific drug targets. Cell Rep. 2013, 5: 216-223. 10.1016/j.celrep.2013.08.028.View ArticlePubMedGoogle Scholar
- de Hoon MJ, Imoto S, Nolan J, Miyano S: Open source clustering software. Bioinformatics. 2004, 20: 1453-1454. 10.1093/bioinformatics/bth078.View ArticlePubMedGoogle Scholar
- Saldanha AJ: Java Treeview--extensible visualization of microarray data. Bioinformatics. 2004, 20: 3246-3248. 10.1093/bioinformatics/bth349.View ArticlePubMedGoogle Scholar
- Metheny-Barlow LJ, Flynn B, van Gijssel HE, Marrogi A, Gerwin BI: Paradoxical effects of platelet-derived growth factor-A overexpression in malignant mesothelioma. Antiproliferative effects in vitro and tumorigenic stimulation in vivo. Am J Respir Cell Mol Biol. 2001, 24: 694-702. 10.1165/ajrcmb.24.6.4334.View ArticlePubMedGoogle Scholar
- Mandhane SN, Shah JH, Bahekar PC, Mehetre SV, Pawar CA, Bagad AS, Chidrewar GU, Rao CT, Rajamannar T: Characterization of anti-inflammatory properties and evidence for no sedation liability for the novel antihistamine SUN-1334H. Int Arch Allergy Immunol. 2010, 151: 56-69. 10.1159/000232571.View ArticlePubMedGoogle Scholar
- Truong-Tran AQ, Ruffin RE, Foster PS, Koskinen AM, Coyle P, Philcox JC, Rofe AM, Zalewski PD: Altered zinc homeostasis and caspase-3 activity in murine allergic airway inflammation. Am J Respir Cell Mol Biol. 2002, 27: 286-296. 10.1165/rcmb.2001-0014OC.View ArticlePubMedGoogle Scholar
- Chambers HF, Kocagoz T, Sipit T, Turner J, Hopewell PC: Activity of amoxicillin/clavulanate in patients with tuberculosis. Clin Infect Dis: an official publication of the Infectious Dis Soc of America. 1998, 26: 874-877. 10.1086/513945.View ArticleGoogle Scholar
- Nadler JP, Berger J, Nord JA, Cofsky R, Saxena M: Amoxicillin-clavulanic acid for treating drug-resistant Mycobacterium tuberculosis. Chest. 1991, 99: 1025-1026. 10.1378/chest.99.4.1025.View ArticlePubMedGoogle Scholar
- Agarwal S: To assess the clinical efficacy of azithromycin and capreomycin in the threatment of multi-drug resistant pulmonary tuberculosis. Chest. 2004, 126: 752S-Google Scholar
- Watt B, Rayner A, Harris G: Comparative activity of azithromycin against clinical isolates of mycobacteria. J Antimicrob Chemother. 1996, 38: 539-542. 10.1093/jac/38.3.539.View 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 credited. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.