A discovery platform for translational research

Transcription

1 A discovery platform for translational research - DisGeNET-RDF&SPARQL - Usage and Modeling Challenges Núria Queralt Rosinach Integrative Biomedical Informatics Group (IBI) Research Programme on Biomedical Informatics (GRIB) Hospital del Mar Research Institute (IMIM) Pompeu Fabra University (UPF) Barcelona

2 Outline How can DisGeNET help your research? DisGeNET Discovery Platform Overview DisGeNET Linked Open Data Introduction RDF-LD Description: Data Model, VoID, Interlinking Implementation Accessibility Documentation Use Cases Querying the DisGeNET-RDF Hands-on

3 How can DisGeNET help your research?

4 Big Questions 4 Big Data -Data Bases -Literature -OMICS -Animal models - Human Biology Understanding Human Diseases Medical Sciences -EMR, EHR, IoT -Imaging -Patient registries -Clinical trials -Epidemiologic studies - PPI Genotype Phenotype Comorbidities Environment (life-style, chemicals, radiation, infections, clinical care intervention, ) DDI

5 Translational Research -Data Bases -Literature -OMICS -Animal models - Molecular Understanding Human Diseases Patient -EMR, EHR, IoT -Imaging -Patient registries -Clinical trials -Epidemiologic studies - Key in Translational Research Genotype Environment Phenotype Decision-making Prevention Diagnosis Therapies Research Discovery

6 Access to Gene-Disease Associations Mental retardation -?- SOX3 SOX3 OMIM:300123; OMIM: ORPHA393; ORPHA90695; ORPHA3157; ORPHA79495; ORPHA67045 Mental Retardation; Panhypopituitarism; 46,XX sex reversal 3 No Data MESH:C538613; MESH:C538613

7 Access to Gene-Disease Associations Mental retardation -?- SOX3 SOX3 OMIM:300123; OMIM: ORPHA393; ORPHA90695; ORPHA3157; ORPHA79495; ORPHA67045 Mental Retardation; Panhypopituitarism; 46,XX sex reversal 3 No Data MESH:C538613; MESH:C Lack of: Normalization Semantic integration Data model harmonization Unified access

8 Knowledge platform on human gene-disease associations (GDAs) Integrates information from expert-curated databases and from the literature (text mining) All disease areas Supporting evidence Analysis tools Piñero et al. DisGeNET: a discovery platform for the dynamical exploration of human diseases and their genes. Database (2015) Vol. 2015: article ID bav028, (2015)

9 Research Questions Which genes are associated to Marfan syndrome? Which disease genes have approved drugs annotated? Which disease genes have differential expression? Which disease genes share a pathway? Is there genetic variation related to the MECP2 and Rett Syndrome association? What evidence supports the association between APP gene and Alzheimer Disease? Which genes and evidence support the comorbidity between Chronic Kidney disease and Diabetes Mellitus, Type 2? Evidence ANALYSIS KNOWLEDGE DISCOVERY ACTIONABLE INFORMATION

10 DisGeNET Discovery Platform Overview

11 DisGeNET Implementation Text mining Biomedical databases Bio-Entity Finder and Relation Extraction Gene-disease associations Gene-disease associations

12 DisGeNET Sources DisGeNET v3.0 GAD LHGDN CURATED PREDICTED LITERATURE

13 Data Integration EVIDENCE DisGeNET ONTOLOGY S = Scurated + Spredicted + Sliterature Score Source Article Gene-Disease Sentence SNP Disease Gene MeSH Class UMLS STY Protein Panther Class Pathway Use of Standards and controlled vocabularies

14 DisGeNET Statistics (May 15th, 2015) Annual Release DisGeNET v3.0 Source Genes Diseases Associations Curated 7,878 6,761 26,522 Predicted 2,557 2,003 9,536 Literature 16,298 11, ,175 All 17,181 14, , % Large volume of information unlocked by text mining the literature

15 Tools for exploration (Onexus) Usage stats (Ago2014-Ago2015): 12,040 users, 22,696 sessions (4:33 min/session) 14,494 downloads (database, Cytoscape plugin, RDF/Nanopubs) DisGeNET used in +20 publications, cited in +60 articles

16 Web Interface SOX3 NCBI:6658 Panhypopituitarism UMLS:C

17 Cytoscape Plugin SOX3 NCBI:6658 Panhypopituitarism UMLS:C

18 DisGeNET Linked Open Data

19 DisGeNET as Linked Open Data RDF and trusty nanopublications URIs: RDF providers or SIO Use of standards (11 ontologies in NCBO) Metadata description (W3C HCLS) Interlinking Bio2RDF Linked Life Data Access Download Data Dump SPARQL Endpoint Faceted Browser Open PHACTS Nanopublication Network Open license Datahub Software BioHackathon Aug 2014

20 DisGeNET-RDF

21 Data Model How to describe an association? Gene associated Disease a) As a property S P O b) As a class Gene Association Disease O P S P O

22 Data Model How to describe an association? Gene associated Disease a) As a property S P O b) As a class Gene Association Disease O P S P O

23 Data Model How to describe an association? Gene associated Disease a) As a property Provenance and Evidence S RDF triples P O b) As a class Gene Association Disease O P S P O

24 Ontology-based integration Data Model DisGeNET Association Type Ontology DisGeNET Standards Shared IDs Standard ontologies rdf:type Gene Association Disease O P S P O

25 Data Model Semantic Annotation: Standard ontologies Prefix Namespace Vocabularies ncit NCI Thesaurus sio SIO up UniProt void VoID foaf FOAF Vocabulary dcterms DCMI Terms rdf RDF rdfs RDF Schema xsd XML Schema owl OWL skos SKOS

26 Data Model Semantic Annotation: Standard ontologies Prefix Namespace Vocabularies ncit NCI Thesaurus sio SIO up UniProt void VoID foaf FOAF Vocabulary dcterms DCMI Terms rdf RDF rdfs RDF Schema xsd XML Schema owl OWL skos SKOS RDF Structure

27 Data Model Semantic Annotation: Standard ontologies Prefix Namespace Vocabularies ncit NCI Thesaurus sio SIO up UniProt void VoID foaf FOAF Vocabulary dcterms DCMI Terms rdf RDF rdfs RDF Schema xsd XML Schema owl OWL skos SKOS Biomedical entities Relationships RDF Structure

28 Data Model Semantic Annotation: Standard ontologies Prefix Namespace Vocabularies ncit NCI Thesaurus sio SIO up UniProt void VoID foaf FOAF Vocabulary dcterms DCMI Terms rdf RDF rdfs RDF Schema xsd XML Schema owl OWL skos SKOS Biomedical entities Relationships Metadata RDF Structure

29 Data Model URIs in DisGeNET: shared, cool & dereferenceable ID Normalization DisGeNET URIs: Estable URIs from primary data providers Identifiers.org Unique association attributes

30 Data Model URIs in DisGeNET: shared, cool & dereferenceable ID Normalization Gene-Disease Association::DisGeNET ID Entity URI Semantics Gene-Disease Association DGNf5cb3969d75871f05a5d5f984f8dfc34 sio:sio_ PubMed article ncit:c47902 Source Score ncbigene:4728_umls:c _association_disgenet Score dctypes:dataset, dcat:distribution ncit:c25338 SNP ncit:c18279

31 Data Model URIs in DisGeNET: shared, cool & dereferenceable ID Normalization Gene::NCBI Gene ID Entity URI Semantics Gene ncit:c16612 HGNC Gene Symbol ncit:c43568 Protein ncit:c17021 Panther Class /PC00211 rdfs:class Pathway ncit:c20633

32 Data Model URIs in DisGeNET: shared, cool & dereferenceable ID Normalization Disease::UMLS Concept Unique Identifier (CUI) Entity URI Semantics Disease C ncit:c7057 MeSH Class rdfs:class UMLS Semantic Type umlssn.owl#t047 rdfs:class Phenotype sio:sio_ Cross References Human Disease Ontology, MesH, OMIM, Orphanet, Decipher, NCIt, ICD9, Human Phenotype Ontology

33 Data Model

34 Data Model /DisGeNET-RDF-Example.ttl (Turtle)

35 Metada Dataset Description DisGeNET-RDF VoID file (Vocabulary of Interlinked Datasets) Gene Disease Association Disease Class Pathway SNP STY PubMed Panther Class Protein DisGeNET-RDF Pathway HGNC Symbol subsets Dataset DisGeNET Database DisGeNET CTD UniProt ClinVar MGD BeFree 1º sources

36 Interlinking DisGeNET -- RDF link -> LOD cloud Dataset 1 Dataset 2 DisGeNET skos:exactmatch UniProt Important in Federated Queries!

37 Interlinking?s skos:exactmatch?o Biomedical Databases and Disease Terminologies NCBI Gene PubMed Orphanet UniProt DisGeNET OMIM dbsnp UMLS DBpedia MeSH

38 DisGeNET as Linked Open Data Interlinking: 4,962,315 RDF links to RDF datasets in the LOD (more statistics)

39 Federated Query Support SPARQL 1.1: SERVICE <sparql endpoint> {} Skos:exactMatch GDA ID Disease ID Gene ID

40 Implementation DisGeNET RDF data, VoID dataset description, and six OWL ontologies loaded into the RDF Store Total number of triples: 24,882,432 (8,5G) LODEStar: SPARQL + LD Browser SPARQL Endpoint Faceted Browser Hardware: Usage Restrictions SPARQL: only SELECT, DESCRIBE, ASK performance opt: Max # of rows per result Max query cost estimation time Max query execution time Security: basic setup

41 Accessibility Download: RDF dump + linksets Faceted Browser SPARQL endpoint EBI::LODEStar SPARQL + Linked Data Browser Open PHACTS APIs

42 Documentation Descriptions RDF Schema Points of access SPARQL query

43 Use Cases What genes are associated to Marfan syndrome? What evidence supports the association between APP gene and Alzheimer Disease? What disease classes are associated with APP gene? Which genes and evidence support the comorbidity between Chronic Kidney disease and Diabetes Mellitus, Type 2? What SNPs are related to the MECP2 and Rett Syndrome association? Which diseases are associated to post-translational modifications type of association? What disease genes are hitted by compounds in ChEMBL? What disease genes have differential expression in Gene Expression Atlas? What disease genes are in WikiPathways? Find compounds (from ChEMBL) that target genes (from DisGeNET) that participate in the same pathway (from WikiPathways)

44 Querying the DisGeNET-RDF

45 SPARQL QUERIES Not easy RDF Schema-aware Performance issues Optimal queries: there is a trade off between the amount of time you spend analyzing and transforming the query and the performance gains of those transformations Technology-dependant crossing a lot of information decrease speed (making the system fails): better local Other approaches on development Q/A based on natural language Linked Data Fragments ElasticSearch

46 Querying DisGeNET SPARQL Queries over DisGeNET data SPARQL query graph

47 Data Model

48 Querying DisGeNET SPARQL Queries over DisGeNET data Minimal Resource Description Graph?label?type rdf:type?subject?rdfsource?comment?title?id SELECT DISTINCT * FROM < WHERE{?subject rdf:type?type ; rdfs:label?label; rdfs:comment?comment ; dcterms:identifier?id ; dcterms:title?title ; void:indataset?rdfsource. } LIMIT 100

49 Querying DisGeNET SPARQL Queries over DisGeNET data Gene-Disease Association Graph sio:sio_ rdf:type gda/dgnf5cb3969d75871f05a5d 5f984f8dfc34 SELECT DISTINCT?gda FROM < WHERE{?gda rdf:type sio:sio_ } LIMIT 100 Which is the sio:sio_ class? For each?gda, show me the?gene and the?disease associated, and the?typeofassociation For each?gda, show me the?gene and the?disease associated, the?paper, and the?sentence description of the relationship in the paper For each?gda show me the?gene,?disease,?source, and the level of?evidence of the association For each gene-disease pair show me the score?value For each?gda show me the?snp Go to the Web and understand and execute Q1.1-Q1.4

50 Querying DisGeNET SPARQL Queries over DisGeNET data Gene Graph Gene rdf:type SELECT DISTINCT?gene FROM < WHERE{?gene rdf:type ncit:c } LIMIT 100

51 Querying DisGeNET SPARQL Queries over DisGeNET data Gene Graph Gene rdf:type SELECT DISTINCT?gene FROM < WHERE{?gene a ncit:c } LIMIT 100 For each?gene show me:?identifier,?name,?genesymbol?protein(s)?panther class(es) and?pantherclassname?pathway(s) and?pathwayname Go to web and understand/execute Q1.5

52 Querying DisGeNET SPARQL Queries over DisGeNET data Disease Graph Disease rdf:type /id/c SELECT DISTINCT?disease FROM < WHERE{?disease a Disease. } LIMIT 100 For the disease < show me: the disease?name, MeSH disease class?label, and the umlssty?title show all cross-references to other disease terminologies Go to the Web and understand/execute Q1.6

53 Querying DisGeNET SPARQL Queries over DisGeNET data Ontology Walking queries Grouping of similar instances Filtering data Query data by classes Ontologies loaded in our RDF triple store: SIO, DO, ORDO, NCIT, HPO, and ECO (OWL) Go to the Web and understand/execute Q1.7and Q1.11

54 Querying DisGeNET SPARQL Queries over DisGeNET data Disease-Phenotype Association Graph (curated from HPO) Disease rdf:type /id/c ?link Disease rdf:type sio: is-manifested-as?phenotype Go to the Web and understand/execute Q1.10 and Q1.12

55 Querying DisGeNET SPARQL Queries over DisGeNET data Disease-Phenotype Association Graph (curated from HPO) Why this model? SELECT DISTINCT?disease count(distinct?hpdisease) as?hpdiseases count(distinct?phenotype) as?phenotypes WHERE {?disease rdf:type ncit:c7057.?disease skos:exactmatch?hpdisease.?hpdisease sio:sio_000341?phenotype. } ORDER BY DESC(?hpdiseases) LIMIT 100 SELECT DISTINCT?disease?hpdisease count(distinct?phenotype) as?phenotypes WHERE {?disease rdf:type ncit:c7057.?disease skos:exactmatch?hpdisease.?hpdisease sio:sio_000341?phenotype. FILTER (?disease = < } GROUP BY?disease?hpdisease

56 Querying DisGeNET + LOD cloud Federated Queries: DisGeNET + external datasets Go to the Web and understand/execute the Federated Queries

57 Use Cases What genes are associated to Marfan syndrome? What evidence supports the association between APP gene and Alzheimer Disease? What disease classes are associated with APP gene? Which genes and evidence support the comorbidity between Chronic Kidney disease and Diabetes Mellitus, Type 2? What SNPs are related to the MECP2 and Rett Syndrome association? Which diseases are associated to post-translational modifications type of association? What disease genes are hitted by compounds in ChEMBL? What disease genes have differential expression in Gene Expression Atlas? What disease genes are in WikiPathways? Find compounds (from ChEMBL) that target genes (from DisGeNET) that participate in the same pathway (from WikiPathways)

58 Acknowledgments IBI Group Alba Gutiérrez-Sacristán Àlex Bravo Janet Piñero Núria Queralt Rosinach Alexia Giannoula Miguel A. Mayer Laura I. Furlong Ferran Sanz PRBB Computational Seminars

59 Thanks for your attention! Questions are welcome