fédération de données et de ConnaissancEs Distribuées en Imagerie BiomédicaLE Interrogation d'entrepôts distribués et hétérogènes

Transcription

1 fédération de données et de ConnaissancEs Distribuées en Imagerie BiomédicaLE Interrogation d'entrepôts distribués et hétérogènes Johan Montagnat Alban Gaignard MI CNRS appel MASTODONS 1

2 Contexte Equipe MODALIS, laboratoire I3S (Sophia Antipolis) Département INS2I du CNRS Génie logiciel et calcul distribué à grande échelle Projet CrEDIBLE Mission pour l'interdisciplinarité du CNRS Appel MASTODONS 2011 (masses de données) Projet reconductible sur 5 ans (d'année en année) Animation de réseau scientifique (cf atelier 2013) CNRS / U. Nice (I3S), INRIA (Sophia), INSERM (Rennes), U. Picardie (MIS), INSA / U. Lyon 1 (CREATIS) MI CNRS appel MASTODONS 2

3 Biomedical data Motivations High heterogeneity: images, clinical data, biomarkers, biology... Increasing amount / number of (open) sources Large-scale medical studies (statistical medical studies, epidemiology...) Need for cross-factors analysis Data (re)analysis opportunities Translational research Linked Data Big Data Centralized approaches encounter limitations Large data volumes to transfer / archive / search Sensitive patient data / complex access control policies Need to adopt uniform data model & format Data is de facto distributed over acquisition centers MI CNRS appel MASTODONS 3

4 Biomedical data mediation & federation Data federation through distributed querying and query rewriting Client Federator (query decomposition, planning & results federation) Remote sub-queries Query-based access to data Site 1 Mediator Site 2 (ETL or query rewriting) Mediator (ETL or query rewriting) Heterogeneous databases schema mediation Medical data & metadata: raw data + models + processing results + models + provenance... MI CNRS appel MASTODONS 4

5 Domain ontology-based federation Medical domain ontology (reference model) Data querying Client Query-based access to data Federator Remote sub-queries Data alignment Mediator Mediator Site 1 Site 2 MI CNRS appel MASTODONS 5

6 Reference ontology 3-levels structure DOLCE foundational ontology, core ontologies, domain ontologies Covering DataSets / Subjects / Studies Data Processing Tools ROIs and ROI Annotations Scientific Measurements Clinical Tests, Scores and instrument-based Assessment Medical context Data provenance... Domain-specific rules Inference abilities Derived relational schema MI CNRS appel MASTODONS 6

7 Reference ontology Particular Endurant Perdurant Dataset acquisition Subjects Languages Medical Action equipment Image formats Centres Investigators Dataset processings Dataset acquisitions Inscription Expression Conceptualization Studies Medical image files Medical image expressions Datasets Examinations MI CNRS appel MASTODONS 7

8 Ontology modules Modularized ontology to improve reuse and lightweightness ONL-MR-DA: MR Dataset Acquisition ONL-DP: Data Processing ONL-MSA: Mental State Assessment OntoVIP: Medical Image Simulation MI CNRS appel MASTODONS 8

9 Data query and federation engine KGRAM (Knowledge Graph Abstract Machine) Semantic query engine: Full support of SPARQL1.1 Generic interface for heterogeneous backends Flexible architecture facilitating different deployment scenarios Mediation interface to access relational data Federated relational schema derived from the ontology MI CNRS appel MASTODONS 9

10 Query federator decoupled from data sources Asynchronous querying of multiple data sources Query planning and parallel querying MI CNRS appel MASTODONS 10

11 KGRAM query processing Q SELECT?name?date WHERE {?x foaf:name?name.?x dbpedia:birthdate?date. FILTER (CONTAINS (?name, 'Bob')) } Asynchronous execution MI CNRS appel MASTODONS 11

12 KGRAM query processing Q SELECT?name?date WHERE {?x foaf:name?name.?x dbpedia:birthdate?date. Q1 FILTER (CONTAINS (?name, 'Bob')) } Q2 Asynchronous execution MI CNRS appel MASTODONS 12

13 KGRAM query processing Q SELECT?name?date WHERE {?x foaf:name?name.?x dbpedia:birthdate?date. Q1 FILTER (CONTAINS (?name, 'Bob')) } Q2 Asynchronous execution Interface KGRAM core Q #1 #2 MI CNRS appel MASTODONS 13

14 KGRAM query processing Q SELECT?name?date WHERE {?x foaf:name?name.?x dbpedia:birthdate?date. Q1 FILTER (CONTAINS (?name, 'Bob')) } Q2 Asynchronous execution Interface KGRAM core Q #1 #2 MI CNRS appel MASTODONS 14

15 KGRAM query processing Q SELECT?name?date WHERE {?x foaf:name?name.?x dbpedia:birthdate?date. Q1 FILTER (CONTAINS (?name, 'Bob')) } Q2 Asynchronous execution Interface KGRAM core #1 Q Q1 #2 MI CNRS appel MASTODONS 15

16 KGRAM query processing Q SELECT?name?date WHERE {?x foaf:name?name.?x dbpedia:birthdate?date. Q1 FILTER (CONTAINS (?name, 'Bob')) } Q2 Asynchronous execution Interface KGRAM core #1 Q Q1 #2 MI CNRS appel MASTODONS 16

17 KGRAM query processing Q SELECT?name?date WHERE {?x foaf:name?name.?x dbpedia:birthdate?date. Q1 FILTER (CONTAINS (?name, 'Bob')) } Q2 Asynchronous execution Interface KGRAM core Q Q1 Q2' #1 #2 MI CNRS appel MASTODONS 17

18 KGRAM query processing Q SELECT?name?date WHERE {?x foaf:name?name.?x dbpedia:birthdate?date. Q1 FILTER (CONTAINS (?name, 'Bob')) } Q2 Asynchronous execution Interface KGRAM core Q Q1 Q2' #1 #2 MI CNRS appel MASTODONS 18

19 KGRAM query processing Q SELECT?name?date WHERE {?x foaf:name?name.?x dbpedia:birthdate?date. Q1 FILTER (CONTAINS (?name, 'Bob')) } Q2 Asynchronous execution Interface KGRAM core #1 Q Q1 Q2' Q2'' #2 MI CNRS appel MASTODONS 19

20 KGRAM query processing Q SELECT?name?date WHERE {?x foaf:name?name.?x dbpedia:birthdate?date. Q1 FILTER (CONTAINS (?name, 'Bob')) } Q2 Asynchronous execution Interface KGRAM core #1 Q Q1 Q2' Q2'' #2 MI CNRS appel MASTODONS 20

21 KGRAM query processing Q SELECT?name?date WHERE {?x foaf:name?name.?x dbpedia:birthdate?date. Q1 FILTER (CONTAINS (?name, 'Bob')) } Q2 Asynchronous execution Interface KGRAM core #1 Q Q1 Q2' Q2'' #2 MI CNRS appel MASTODONS 21

22 KGRAM query processing Q SELECT?name?date WHERE {?x foaf:name?name.?x dbpedia:birthdate?date. Q1 FILTER (CONTAINS (?name, 'Bob')) } Q2 Asynchronous execution Interface KGRAM core #1 Q Q1 Q2' Q2'' #2 MI CNRS appel MASTODONS 22

23 Performance results Heterogeneous (relational / semantic) stores querying FedBench standard benchmark MI CNRS appel MASTODONS 23

24 Deployment Customizable for different deployment scenarios MI CNRS appel MASTODONS 24

25 Conclusions Query-based data federation Using semantic web standards (SPARQL, RDF) Emphasis on query language expressivity Ontology-based Reference model for data alignment and query terms Currently deployed at medium scale Broad applicability (standards compliance) given that ontologies are available Query optimization work on-going MI CNRS appel MASTODONS 25