ChemDIS: a chemical–disease inference system based on chemical–protein interactions
Journal of Cheminformatics volume 7, Article number: 25 (2015)
The characterization of toxicities associated with environmental and industrial chemicals is required for risk assessment. However, we lack the toxicological data for a large portion of chemicals due to the high cost of experiments for a huge number of chemicals. The development of computational methods for identifying potential risks associated with chemicals is desirable for generating testable hypothesis to accelerate the hazard identification process.
A chemical–disease inference system named ChemDIS was developed to facilitate hazard identification for chemicals. The chemical–protein interactions from a large database STITCH and protein–disease relationship from disease ontology and disease ontology lite were utilized for chemical–protein–disease inferences. Tools with user-friendly interfaces for enrichment analysis of functions, pathways and diseases were implemented and integrated into ChemDIS. An analysis on maleic acid and sibutramine showed that ChemDIS could be a useful tool for the identification of potential functions, pathways and diseases affected by poorly characterized chemicals.
ChemDIS is an integrated chemical–disease inference system for poorly characterized chemicals with potentially affected functions and pathways for experimental validation. ChemDIS server is freely accessible at http://cwtung.kmu.edu.tw/chemdis.
Humans are exposed to thousands of chemicals in everyday life. Nevertheless, the toxicological data required for risk assessment are largely unknown for a large portion of chemicals. Instead of applying in vitro or in vivo experiments directly that are expensive and time-consuming, the computational integration of existing toxicogenomics information for the inference of potential toxicities and pathways could largely accelerate the process of risk assessment.
For the integration of toxicogenomics information, the Comparative Toxicogenomics Database (CTD) was constructed by curating chemical–gene/protein interactions from more than 100,000 selected articles for a decade [1, 2]. The chemical–gene–disease associations could be inferred by combining chemical–gene interactions with gene–diseases associations. CTD consisting of high-confidence chemical–gene interactions is a useful resource for studying chemical-induced diseases. Please note that the inferred associations could be either therapeutic or toxic effects. While the analysis of chemical–gene/protein interactions could be useful for narrowing down potentially affected diseases, the interactions alone can not be used to determine whether a chemical induces therapeutic or toxic effects due to the complex nature of biological systems involving various interaction types. Experiments should be subsequently applied to determine which effects are associated with a given chemical. In spite of the limitation, the inference analysis is capable of identifying a small subset of potentially affected diseases with interacting genes/proteins for experimental validation that greatly accelerates the hazard identification process. Traditional bioassays are usually designed for a few specific toxicological and pharmacological endpoints. The integrated analysis of interactions reported from individual toxicology and pharmacology studies is of great importance giving systematic effects that may not be easily observed from the individual studies. However, for poorly characterized chemicals, only a few interacting genes were curated in CTD making the inference of potential diseases impossible.
Instead of analysis of enriched diseases from all interacting genes, ChemProt  and HExpoChem  focused on analyzing diseases for each chemical-interacting gene/protein based on protein–protein interactions. Although the one-by-one analysis of diseases for each gene could be helpful for studying chemical-induced diseases, a systematic enrichment analysis based on all interacting genes/proteins could provide overall effects that are more easily interpretable.
Recently, a computational inference approach was proposed to identify potential diseases associated with maleic acid, a poorly characterized chemical with only one gene curated in CTD database . The utilization of chemical–protein interaction data from STITCH 3.1, one of the largest chemical–protein interaction databases , enabled the inferences of functions, pathways and diseases affected by maleic acid. The approach is potentially useful for the identification of diseases associated with poorly characterized chemicals.
In order to facilitate the inferences of functions, pathways and diseases affected by various environmental and industrial chemicals, a comprehensive resource named ChemDIS was constructed by integrating the chemical–protein interactions in human from STITCH database and enrichment analysis tools. The newly published STITCH 4 with 45% more high-confidence interactions than its previous version  was integrated that enlarged the applicability domain of ChemDIS to poorly characterized chemicals. Tools for the enrichment analysis of gene ontology (GO) terms , pathways (KEGG  and Reactome ), disease ontology (DO)  and disease ontology lite (DOLite)  were implemented and integrated in ChemDIS.
The usefulness of ChemDIS for poorly characterized chemicals was demonstrated by an analysis of maleic acid and sibutramine. ChemDIS successfully inferred kidney diseases that were reported in a safety assessment of maleic acid  but not identified in our previous study . In addition, newly identified immune system and infectious diseases provide directions for future studies. For the analysis of sibutramine, the previously reported adverse effects including hypertension, myocardial infarction, heart disease, anorexia nervosa and bipolar disorder [14–17] were also successfully identified by ChemDIS. ChemDIS with user-friendly interfaces is expected to be a useful server for identifying potential risks associated with poorly characterized chemicals.
Figure 1 shows the system flow of ChemDIS. The chemical information of structures and physicochemical properties was downloaded from PubChem . OpenBabel  was applied to represent the 2D structures of chemicals. Chemical–protein interaction data were retrieved from STITCH database, an aggregated database of interactions from several interaction databases , such as CTD , ChEMBL , DrugBank , Kyoto Encyclopedia of Genes and Genomes (KEGG)  and Reactome . The aggregated interaction data were also shown to be useful for predicting non-genotoxic hepatocarcinogenicity . Both STITCH databases of versions 4 and 3.1 were integrated into ChemDIS connecting over 300,000 chemicals and 19,489 and 16,973 human proteins, respectively. The combined scores obtained from STITCH were used for filtering interacting proteins with three confidence levels of high, medium and low.
Enrichment analysis tools were implemented and integrated into ChemDIS for analyzing functions, pathways and diseases affected by a given chemical. For enriched functions, clusterProfiler  will be applied to analyze enriched gene ontology (GO) terms for molecular function, biological process and cellular component. The enriched KEGG  and Reactome  pathways will be analyzed using clusterProfiler and ReactomePA , respectively. For inferring diseases affected by a given chemical, enriched DO  and DOLite  terms will be analyzed using DOSE package . DOLite is a simplified vocabulary list from DO, a standardized ontology connecting human proteins to diseases. All the enrichment analyses are based on hypergeometric tests with the Benjamini–Hochberg approach  for multiple testing correction. Enriched terms with a corrected p value <0.05 will be identified.
Results and discussion
ChemDIS provides a unique resource for inferring functions, pathways and diseases associated with chemicals based on chemical–protein interactions. Figure 2 shows the web-interface of ChemDIS equipped with a quick search tool and a hyperlink to advanced search tool. The quick search tool utilizes default parameters of 0.15 for interaction score and 4 for database version. Given a chemical queried by users, its basic structure and property information including chemical 2D structure, hydrogen-bond acceptor, hydrogen-bond donor, IUPAC name, INCHI, INCHIKEY, molecular formula, molecular weight, canonical SMILES, isomeric SMILES and topological polar surface area (TPSA) is available at ChemDIS. To give insights into the functions and pathways affected by chemicals, built-in functions are available for analyzing enriched GO terms and pathways (KEGG and Reactome).
Diseases associated with chemicals will be inferred from their interacting proteins based on DO and DOLite. The utilization of standardized DO terms integrated from multiple ontology sources  is expected to provide comprehensive analysis results, while DOLite terms offer simplified disease terms that are more interpretable. Hyperlinks to external databases are available for detailed information of chemicals, proteins, genes, GO, pathways and DO. Result tables are sortable by clicking the header of tables with search functions for filtering results. All analysis results generated from ChemDIS are downloadable.
Due to the dependence of ChemDIS on chemical–protein interactions, the number of interactions for a given chemical determines its applicability domain. For chemicals with more than or equal to 30 interacting human proteins, 14,831 chemicals can be analyzed at ChemDIS, compared with 1,190 chemicals and 2,097 chemicals for CTD without and with the incorporation of cross-species interactions, respectively (Access date: April 5, 2015). Table 1 shows the detailed comparison of ChemDIS and CTD. ChemDIS utilizing chemical–protein interactions from the large database STITCH 4 enables the inference of potential risks for a wide range of chemicals.
Case study of maleic acid
As a case study, the potential risks of maleic acid on human health were reanalyzed using ChemDIS and compared with our previous study . Based on STITCH 4, 36 genes mapped from maleic acid-interacting proteins were identified using the keyword ‘maleate’, a synonym of maleic acid, and default threshold 0.15 for interaction score that all interacting proteins will be utilized for the following analysis. Hyperlinks to Ensembl  protein database and NCBI Gene database were also available for detailed information.
Both neuronal system and metabolism were identified to be potentially affected by maleic acid from GO and pathway enrichment analyses that were consistent with our previous report. Hyperlinks to external databases of QuickGO [30, 31], KEGG and Reactome were also available at ChemDIS. DO enrichment analysis confirmed that disease of mental health, nervous system disease and disease of metabolism could be potentially associated with maleic acid. The identification of cardiovascular diseases was also consistent with our previous study.
A snapshot of DO enrichment analysis for maleic acid is shown in Figure 3. The gene ratio indicates the ratio between the number of interacting genes associated with a DO term and the number of interacting genes mapped to DO terms. The ratio between the number of genes associated with a DO term and the number of genes mapped to DO terms is represented as background ratio (Bg Ratio). The p value and adjusted p value are calculated based on the hypergeometric test without and with multiple test correction, respectively. The q value is a measure of false discovery rate .
Newly identified diseases included immune system, kidney and infectious diseases. Notably, kidney diseases inferred by ChemDIS has been shown in experimental animals that our previous study failed to identify . ChemDIS successfully identified known diseases associated with maleic acid including kidney , behavioral and gastrointestinal diseases . DOLite enrichment analysis showed that hypertension could be associated with maleic acid. The detailed analysis results for maleic acid is available in Additional file 1: Table S1. In summary, ChemDIS identified 3 and 29 DO terms for known and newly identified diseases from 8,727 DO terms, respectively. In addition, a newly identified DOLite term of hypertension was identified. The analysis results provide future directions of toxicological research on maleic acid. For CTD, 1 and 56 disease terms were identified for known and newly identified diseases, respectively. Please note that the inference from CTD was based on only one gene giving low-scoring diseases without sufficient information for further experimental validation.
Case study of sibutramine
In addition to the maleic acid with less known associated diseases, a withdrawal drug sibutramine was used to evaluate the ability of ChemDIS to identify known associated diseases. Similar to maleic acid, there is only five genes curated in CTD database giving only partial interaction information making the disease inference difficult. Sibutramine is originally indicated for the management of obesity and has been withdrawn from the market due to the concern of cardiotoxicity [15, 16]. The reported adverse effects associated with sibutramine include symptoms of cardiovascular, nervous and gastrointestinal system diseases and disease of mental health. Hypertension, myocardial infarction, arrhythmias, tachycardia, stroke, bipolar disorder, headache, insomnia, constipation, anorexia nervosa and sexual dysfunction have been reported to be associated with sibutramine [14–17, 34–36].
ChemDIS identified 44 genes mapped from sibutramine-interacting proteins based on STITCH 4. The DO terms of cardiovascular system disease, hypertension, myocardial infarction and heart disease were successfully identified. The enriched DO term of heart disease accounts the arrhythmias, tachycardia and stroke. ChemDIS performs well for identifying known sibutramine-induced cardiotoxicity. The corresponding DO terms for nervous system disease were also identified including nervous system disease and bipolar disorder. The enriched DO term of nervous system disease implies the symptoms of headache and insomnia. For constipation, the DO term of gastrointestinal system disease was identified. For the disease of mental health, the corresponding DO terms of disease of mental health and anorexia nervosa were identified accounting the adverse effects of sexual dysfunction and anorexia nervosa. A DOLite analysis also successfully identified hypertension and anorexia nervosa. As the interaction data grows, the inferred diseases could be more precise. In addition to the adverse effects, the desired therapeutic effects were also identified as DO terms of obesity, fatty liver disease, overnutrition, nutrition disease and eating disorder and the DOLite term of obesity [37, 38]. The detailed analysis results for sibutramine is available in Additional file 2: Table S2.
Generally, ChemDIS identified 103 DO terms and 7 DOLite terms from a large pool of disease terms that largely help the prioritization of potentially associated diseases. Among the 103 identified DO terms, 10 and 5 terms are consistent with previously reported adverse and therapeutic effects, respectively. For the 7 inferred DOLite terms, there are 2 and 1 terms corresponding to known adverse and therapeutic effects. Newly identified associations include the remaining 88 and 4 terms for DO and DOLite, respectively. For CTD, most of the identified 57 disease terms were low-scoring associations that the average number of genes used for each inference is only 1.12. While 5 and 2 terms from CTD analysis were consistent with the previously reported adverse and therapeutic effects, respectively, it is difficult to experimentally validate the results without sufficient information.
ChemDIS is an integrated chemical–disease inference system with a user-friendly interface. Benefit from the integration of the large STITCH database, ChemDIS is expected to be helpful for inferring potential diseases associated with poorly characterized chemicals. The integration of analysis tools enabled the identification of affected functions and pathways that can be further studied experimentally. The analysis of maleic acid and sibutramine demonstrated the capability of ChemDIS for identifying a small number of potential affected diseases from the large pool of disease terms. To further improve the applicability of ChemDIS to chemicals without sufficient interaction data, future works could be the implementation of pharmacophore- and docking-based target identification methods such as PharmMapper [39, 40] and PDTD , respectively, and incorporation of predicted targets for enrichment analysis.
Availability and requirements
ChemDIS is freely available at http://cwtung.kmu.edu.tw/chemdis without restrictions for academic use.
Davis AP, Grondin CJ, Lennon-Hopkins K, Saraceni-Richards C, Sciaky D, King BL et al (2014) The Comparative Toxicogenomics Database’s 10th year anniversary: update 2015. Nucleic Acids Res 43(Database issue):D914–D920
Davis AP, Wiegers TC, Roberts PM, King BL, Lay JM, Lennon-Hopkins K et al (2013) A CTD–Pfizer collaboration: manual curation of 88,000 scientific articles text mined for drug-disease and drug–phenotype interactions. Database (Oxford) 2013:bat080
Kim Kjaerulff S, Wich L, Kringelum J, Jacobsen UP, Kouskoumvekaki I, Audouze K et al (2013) ChemProt-2.0: visual navigation in a disease chemical biology database. Nucleic Acids Res 41(Database issue):D464–D469
Taboureau O, Jacobsen UP, Kalhauge C, Edsgard D, Rigina O, Gupta R et al (2013) HExpoChem: a systems biology resource to explore human exposure to chemicals. Bioinformatics 29(9):1231–1232
Lin YC, Wang CC, Tung CW (2014) An in silico toxicogenomics approach for inferring potential diseases associated with maleic acid. Chem Biol Interact 223C:38–44
Kuhn M, Szklarczyk D, Franceschini A, von Mering C, Jensen LJ, Bork P (2012) STITCH 3: zooming in on protein–chemical interactions. Nucleic Acids Res 40(Database issue):D876–D880
Kuhn M, Szklarczyk D, Pletscher-Frankild S, Blicher TH, von Mering C, Jensen LJ et al (2014) STITCH 4: integration of protein–chemical interactions with user data. Nucleic Acids Res 42(Database issue):D401–D407
Ashburner M, Ball CA, Blake JA, Botstein D, Butler H, Cherry JM et al (2000) Gene ontology: tool for the unification of biology. The Gene Ontology Consortium. Nat Genet 25(1):25–29
Kanehisa M, Goto S, Sato Y, Kawashima M, Furumichi M, Tanabe M (2014) Data, information, knowledge and principle: back to metabolism in KEGG. Nucleic Acids Res 42(Database issue):D199–D205
Croft D, Mundo AF, Haw R, Milacic M, Weiser J, Wu G et al (2014) The reactome pathway knowledgebase. Nucleic Acids Res 42(Database issue):D472–D477
Kibbe WA, Arze C, Felix V, Mitraka E, Bolton E, Fu G et al (2014) Disease Ontology 2015 update: an expanded and updated database of human diseases for linking biomedical knowledge through disease data. Nucleic Acids Res 43(Database issue):D1071–D1078
Du P, Feng G, Flatow J, Song J, Holko M, Kibbe WA et al (2009) From disease ontology to disease-ontology lite: statistical methods to adapt a general-purpose ontology for the test of gene-ontology associations. Bioinformatics 25(12):i63–i68
Cosmetic Ingredient Review Expert P (2007) Final report on the safety assessment of maleic acid. Int J Toxicol 26(Suppl 2):125–130
Luque CA, Rey JA (1999) Sibutramine: a serotonin-norepinephrine reuptake-inhibitor for the treatment of obesity. Ann Pharmacother 33(9):968–978
Scheen AJ (2011) Sibutramine on cardiovascular outcome. Diabetes Care 34(Suppl 2):S114–S119
Yim KM, Ng HW, Chan CK, Yip G, Lau FL (2008) Sibutramine-induced acute myocardial infarction in a young lady. Clin Toxicol (Phila) 46(9):877–879
Waszkiewicz N, Zalewska-Szajda B, Szajda SD, Simonienko K, Zalewska A, Szulc A et al (2012) Sibutramine-induced mania as the first manifestation of bipolar disorder. BMC Psychiatry 12:43
Yet Another DataTables Column Filter. https://github.com/vedmack/yadcf
Bolton EE, Wang Y, Thiessen PA, Bryant SH (2008) Chapter 12—PubChem: integrated platform of small molecules and niological activities. In: Ralph AW, David CS (eds) Annual reports in computational chemistry, vol 4. Elsevier, Amsterdam, pp 217–241
O’Boyle NM, Banck M, James CA, Morley C, Vandermeersch T, Hutchison GR (2011) Open Babel: an open chemical toolbox. J Cheminform 3:33
Gaulton A, Bellis LJ, Bento AP, Chambers J, Davies M, Hersey A et al (2012) ChEMBL: a large-scale bioactivity database for drug discovery. Nucleic Acids Res 40(Database issue):D1100–D1107
Law V, Knox C, Djoumbou Y, Jewison T, Guo AC, Liu Y et al (2014) DrugBank 4.0: shedding new light on drug metabolism. Nucleic Acids Res 42(Database issue):D1091–D1097
Tung CW, Jheng JL (2014) Interpretable prediction of non-genotoxic hepatocarcinogenic chemicals. Neurocomputing 145:68–74
Yu G, Wang LG, Han Y, He QY (2012) clusterProfiler: an R package for comparing biological themes among gene clusters. OMICS 16(5):284–287
Yu G (2012) ReactomePA: reactome pathway analysis. R package version 1.12.1
Yu G, Wang L-G, Yan G-R, He Q-Y (2014) DOSE: an R/Bioconductor package for disease ontology semantic and enrichment analysis. Bioinformatics 31(4):608–609
Benjamini Y, Hochberg Y (1995) Controlling the false discovery rate: a practical and powerful approach to multiple testing. J R Stat Soc B 57(1):289–300
Cunningham F, Amode MR, Barrell D, Beal K, Billis K, Brent S et al (2014) Ensembl 2015. Nucleic Acids Res 43(Database issue):D662–D669
Huntley RP, Binns D, Dimmer E, Barrell D, O’Donovan C, Apweiler R (2009) QuickGO: a user tutorial for the web-based gene ontology browser. Database (Oxford) 2009:bap010
Binns D, Dimmer E, Huntley R, Barrell D, O’Donovan C, Apweiler R (2009) QuickGO: a web-based tool for gene ontology searching. Bioinformatics 25(22):3045–3046
Storey JD (2002) A direct approach to false discovery rates. J R Stat Soc B 64(3):479–498
BIOFAX (1970) Industrial Bio-Test Laboratories, Inc., Data Sheets, vol 7-4
Nojimoto FD, Piffer RC, Kiguti LR, Lameu C, de Camargo AC, Pereira OC et al (2009) Multiple effects of sibutramine on ejaculation and on vas deferens and seminal vesicle contractility. Toxicol Appl Pharmacol 239(3):233–240
Siebenhofer A, Jeitler K, Horvath K, Berghold A, Siering U, Semlitsch T (2013) Long-term effects of weight-reducing drugs in hypertensive patients. Cochrane Database Syst Rev 3:CD007654
Nisoli E, Carruba MO (2000) An assessment of the safety and efficacy of sibutramine, an anti-obesity drug with a novel mechanism of action. Obes Rev 1(2):127–139
Sabuncu T, Nazligul Y, Karaoglanoglu M, Ucar E, Kilic FB (2003) The effects of sibutramine and orlistat on the ultrasonographic findings, insulin resistance and liver enzyme levels in obese patients with non-alcoholic steatohepatitis. Rom J Gastroenterol 12(3):189–192
Wilfley DE, Crow SJ, Hudson JI, Mitchell JE, Berkowitz RI, Blakesley V et al (2008) Efficacy of sibutramine for the treatment of binge eating disorder: a randomized multicenter placebo-controlled double-blind study. Am J Psychiatry 165(1):51–58
Wang X, Chen H, Yang F, Gong J, Li S, Pei J et al (2014) iDrug: a web-accessible and interactive drug discovery and design platform. J Cheminform 6:28
Liu X, Ouyang S, Yu B, Liu Y, Huang K, Gong J et al (2010) PharmMapper server: a web server for potential drug target identification using pharmacophore mapping approach. Nucleic Acids Res 38(Web Server issue):W609–W614
Gao Z, Li H, Zhang H, Liu X, Kang L, Luo X et al (2008) PDTD: a web-accessible protein database for drug target identification. BMC Bioinform 9:104
CWT implemented the program, analyzed the data and wrote the manuscript.
The author thanks Dr. Chia-Chi Wang for proofreading this article. This work was supported by Kaohsiung Medical University Research Foundation (KMU-M104010, KMU-TP103A32), NSYSU-KMU Joint Research Project (NSYSUKMU104-I01-2) and National Health Research Institutes (EH-103-PP-09).
Compliance with ethical guidelines
Competing interests The author declares that he has no competing interests.
About this article
Cite this article
Tung, CW. ChemDIS: a chemical–disease inference system based on chemical–protein interactions. J Cheminform 7, 25 (2015). https://doi.org/10.1186/s13321-015-0077-3