EMBL-EBI Industry Programme Workshop, 26th to 27th November Data Infrastructure for Omics-based Chemical Safety.

Transcription

1 EMBL-EBI Industry Programme Workshop, 26th to 27th November Data Infrastructure for Omics-based Chemical Safety Danyel Jennen

2 The systems toxicology approach Cf. Waters & Fostel. Toxicogenomics and systems toxicology: aims and prospects. Nature Rev Gen 5 (2004)

3 The systems toxicology approach Iterative development of systems models Cf. Fisher & Henzinger. Executable cell biology. Nature Biotech 25 (2007)

4 Modeling MULTISCALE MODELING EXPERIMENTAL Computational chemistry Molecular pathway modeling Physiological modeling in vitro 3D assays Patients Omics analyses Functional analyses Samples Analyses Model construction Data warehousing Data quality control Data Integrated statistics Adjust model PREDICTIVE COMPARISON Suggest new experiments

5 ASAT Knowledge Base ASAT = Assuring Safety without Animal Testing 1 year project in 2011

6 Objectives (A)To define what kind of information is needed and available to translate mechanistic human disease data towards an integrated non animal testing strategy, with liver carcinogenesis and sensitization as endpoints (B)To define if currently available public data from the in vitro toxicological and human disease domain (A) can be integrated (C)To define what the minimal data requirements and data quality need to be to establish such an integration (D)To use data from human disease and in vitro domain to built a relational database: ASAT Knowledge Base

7 Main concept

8 Metadata Compound module HCC Sensitizers Overlap ConnectivityDB ToxProfiler PathVisio Omics data Raw Normalized QC Disease module Gene sets Pathway s/gene sets

9 Array collection

10 Quality control

11 Disease data 24 HCC genesets 4 sensitization genesets

12 Compound information Chemical description Name, Cas number, Abbreviation, synonyms, Formulae In vitro assay Ames, MLA, MN, CA, SCE, UDS Compound In vivo assay MN, CA, SCE, UDS, TRA Carcinogenicity iarc, NTP, CPDB, Effect on animals Other info Mode of action, uses, Supplier Endpoint Compounds Liver 327 Skin 118 Both 15

13 in vitro testing In vivo testing Carcinogenicity Magkoufopoulou Tsamou et al. Performance et al. A transcriptomics-based of in vitro γh2ax assay in vitro in HepG2 assay cells for predicting to in vivo chemical genotoxicity. genotoxicity Mutagenesis in vivo. (2012) Carcinogenesis (2012) Training set 34 compounds Test set 28 compounds Accuracy 89% Sensitivity 91% Specificity 87%

14 ToxProfiler ToxProfiler uses various gene-sets as input Input: 2 log ratio of treatment over corresponding control

15 2-AAF 4-AAF

16 2-AAF 4-AAF

17

18 What s next? Update available data Add Connectivity mapping Extend ASAT Knowledge base pharmacokinetics data follow-up project for 2 years

19 DECO Data-integration for Endpoints, Cheminformatics and Omics. 2 year project March 2012 February 2014

20 Objectives DECO (A)To develop a transparent framework that improves the prediction of the repeated dose toxicity i.e. hepatotoxicity of new chemicals by integrating chemoinformatic data with biological information from omics and HTS technologies (B)To generate a high-quality database that contains for a large number of chemicals, its chemical properties, in vitro omics, in vivo omics, HTS data, and (in vivo) toxicity test (C)To develop and evaluate expert rules, useful for industry as well as regulators

21 Project Organization DECO WP2 Knowledge base with ready to use data Pathways, gene sets, metabolite sets QC, Preprocess, Normalization, Summarization WP1 Chemoinformatics data Toxicity data Metabonomics data Transcriptomics data Data sources: BASF, ToxRefDB, TG-Gates, DrugMatrix, literature BASF TG-Gates, UM, DrugMatrix WP3 Chemoinformatics, Bioinformatics Validation compounds HTS data ToxCast Integrated procedure for hepatotoxicity prediction

22 Selection of data sources and compounds The list of data sources is: 1.The BASF database for metabonomics on plasma from rat in vivo repeated dose studies. In total 386 compounds. 2.DrugMatrix for rat liver transcriptomics, pathology and clinical chemistry data. In total 200 compounds. 3.OPEN TG-GATES for rat liver transcriptomics, pathology and clinical chemistry data (only the repeated dose studies). In total 143 compounds. 4.OPEN TG-GATES for rat and human primary hepatocytes transcriptomics data. In total 138 compounds. 5.ToxCast for data from high throughput screening studies. In total 318 compounds with pathology data. Possibly this can be extended, when data from ToxCast-II become available. 6.The UM database for transcriptomics in human HepG2 cells. In total 122 compounds DECO

23 Focus Groups for integration of chemoinformatics with omics DECO The selection of compounds for four classes: similar chemical structure, similar toxicity profile dissimilar chemical structure, similar toxicity profile similar chemical structure, dissimilar toxicity profile dissimilar chemical structure, dissimilar toxicity profile

24 Calculating similarity values DECO Different datasets: Chemical structure: Tanimoto Presence of certain groups in chemical structures: fingerprint Pathology: Pearson Correlation Coefficient Liver pathology terms per compound Only highest doses and longest exposure times Clinical Chemistry: Pearson Correlation Coefficient ~ 20 clinical chemistry parameters with varying ranges Normalization: dividing by average Only highest doses and longest exposure times

25 Some examples DECO Example of similarity matrix: Heatmap of correlations based on clinical chemistry for 583 compounds in DrugMatrix Comparison of similarity scores for compounds from TG-GATES. Structural similar is not equal to toxicological similar! Will inclusion of omics data improve Toxicity prediction?

26 What s next? DECO Add Omics data Calculate similarity values for omics data Develop prediction tool by integrating data Prediction tool can be added to Cefic s tool box

27 dixa Data Infrastructure for Chemical Safety 3 year project October 2011 September 2014

28 Main Objectives To further develop and adopt a robust and sustainable service infrastructure (e.g. data infrastructure and e- science environment) for harbouring multiplexed data sets as produced by past, current and future EU research projects on developing non-animal tests for predicting chemical safety as conducted by the research community of toxicogenomics To link this with other research communities maintaining globally available chemical/toxicological data bases and data bases on molecular data of human disease.

29 dixa s research hypothesis The main scientific concept which has initiated the dixa proposal, is to create a large public data infrastructure of genomic signatures of drugs, industrial chemicals and cosmetics, with their corresponding meta data to develop pattern-matching bioinformatics and biostatistics tools to detect similarities among these signatures, in order to describe all biological states induced with a chemical exposure, in terms of genomic signatures relevant for the human situation in vivo.

30 Creating dixa s data infrastructure * by integrating TGX data from FP6/FP7 projects * by linking with chemical/mol. medicine data bases * thus enabling cross-platform, cross-study integrations

31 The way forward: integration of transcriptomics, proteomics, metabonomics with epigenetics and μrna and bioinformatics in predictive toxicology DNA methylation microrna

32 Organisational structure

33 -omics data and metadata Metadata harmonization Validation, quality control modelling analysis Full walk-through from data sources to data consumers ChEBI, ChEMBL, ToxCast, PubChem,... KEGG, OMIM,... Toxicogenomics project repositories carcino GENOMIC S PREDTO X... Public repositories ISAcreato rconfig templates ISAcreato r nomenclature, ontologies Chemicals database portal Human disease database dixa warehouse dixa website risata b Array Expres s PRIDE... Genedata analysis workflows Assemble Integrate Query

34 WP3 isatools

35 WP3 Isa software suite

36 WP3 CarcinoGENOMICS Liver Kidney microarray NMR microarray NMR

37 Expanding dixa s toxicogenomics db-1

38 Expanding dixa s toxicogenomics db-2 CEBS Theme 3: Towards global connections in chemical safety What specific challenges will arise in linking to additional databases held by US researchers? Will dixa s service-based mechanisms also work for US dbs, or must they be adapted? What steps can we take now towards common metadata or data schema patterns for global chemical safety databases?

39 WP4 Chemicals database portal

40 Expanding dixa s chemoinformatics db dixa has identified the OECD echemportal as ideal starting point for the cheminformatics portal. OECD has agreed that if constraints laid out in the architecture document, are respected, the 25 participants will not object to their being visible also through the dixa WP4 portal The following data sources ("participants" in OECD echemportal jargon) are therefore candidates also for dixa s chemoinformatics.

41 WP5 Human diseases database

42 WP5 Bottleneck Human data are scarce in comparison to animal data In vitro data focus on acute toxicity, in vivo data are mostly disease related HCC data are over represented Diseases are linked

43 WP5 One disease influences another? HBV fatty liver (steatosis) HCV? chronic (necro) inflammation / fibrosis? HBV HCC fold increased risk alcohol cirrhosis HCC HCV HCC fold increased risk aflatoxin B1

44 WP5 Expanding dixa s clinical db

45 WP8 WP9 Integrative data analysis Computational modeling Newly developed pathways based on literature research Pathways based on NR signaling (AhR, CAR, ER, FXR, etc.) also relevant in hepatotoxic mechanisms Developed by Rianne Fijten Validation of developed pathways with data from TG- GATES shows they are indeed relevant

46 WP8 Meta-pathway of NR pathways: Validation PPAR-α with Wy rat in vivo PPAR-α

47 WP8 Meta-pathway of NR pathways: Validation PPAR-α with Wy rat in vitro PPAR-α

48 WP8 Meta-pathway of NR pathways: Validation PPAR-α with Wy human in vitro PPAR-α

49 WP9 ConsensusPathDB

50 WP9 Creating systems toxicology models Identification of perturbed signaling pathways of toxicity

51 Overall, personalized safety constitutes a major challenge towards handling big data

52 Main conclusions: Where the innovative aspect of e-infrastructures and their role in promoting human safety science and innovation is acknowledged, the relationship between e- Infrastructures and human safety public/private actors needs to be reinforced through generating Joint Research Initiatives. These Joint Research Initiatives should address developing biology-specific middleware linking Safety Science / Systems Toxicology effectively to e-infrastructures. This should obviously be based on the availability of existing middleware/tools. Developing of globally harmonized standards will advance the sustainable integrated knowledge/data/research dixa infrastructure Where effective communication between public e-infrastructures and research and industrial actors is required, dissemination towards applicability across several domains and early involvement of all stakeholders (industry, regulatory bodies, NGOs) is needed dixa fits in the current EC policy considerations because dixa is engaged in worldwide collaboration and data sharing and already acts as a catalyst for safeguarding data and add value to present data in the current context (e.g. past investments). dixa may thus serve as a model example for the working collaboration between ICT providers and domain-specific projects, and develop best practices for both technical solutions & services, also for demonstrating how people can effectively work together.

53 Acknowledgement QUESTIONS