The Bio-Lark Project

An overview of the Biological Large Knowledge Collider (Bio-LarK) and the motivation behind it.

Transcript

1. Andreas Zankl MD, FMH, FRACP Sydney Children’s Hospital (Westmead) University of Sydney The BioLarK Platform
2. Skeletal Dysplasias • genetic disorders affecting skeletal development • rare to ultra-rare (1:10’000 - a few cases worldwide) • similar to neuromuscular or rare renal disorders
3. Number of bone dysplasias 1878 Description of “Achondroplasia” 1929 Description of “Morquio Syndrome” 1970 First International Nosology of Constitutional Diseases of Bone 2010 7th Revision of the Nosology: > 450 disorders 1970 1929 1878 2010
7. The Skeletome Project Disease Gene Mouse Patient Pathway Drug Phenotype similar Patients Expression
 profiles Metabolites
8. Challenges • many different data sources • lack of standardisation • particularly true for phenotype data
9. The Semantic Web • an extension of the WWW • a W3C standard • the WWW connects web pages,
 the Semantic Web connects ‘things’ • URIs instead of URLs • RDF/OWL instead of hyperlinks • ‘web of knowledge’
10. Semantic Web Basics Patient X has_clinical_feature short stature RDF triple URI URI RDF/OWL
11. Patient X short stature has clinical feature has diagnosis Achondroplasia large head has clinical feature has clinical feature has clinical feature Patient Y FGFR3 has mutation in
12. Patient X short stature has clinical feature has diagnosis Achondroplasia large head has clinical feature has clinical feature has clinical feature Patient Y FGFR3 has mutation in
13. Patient X short stature has clinical feature has diagnosis Achondroplasia large head has clinical feature has clinical feature has clinical feature Patient Y FGFR3 has mutation in Ensembl has info OMIM has info
14. Semantic Web Technologies
15. Step 1: Modelling Domain Knowledge Disease Gene Patient Phenotype
16. Turning the ISDS Nosology into an Ontology ISDS Nosology • listing of all known bone dysplasias • grouped by clinical, radiographic and genetic features • developed by international group of experts • updated every 4 years • published in table form in a medical journal
17. 2010 Nosology TABLE I. Group/name of disorder Inheritance MIM No. Locus Gene Protein Notes 1. FGFR3 chondrodysplasia group Thanatophoric dysplasia type 1 (TD1) AD 187600 4p16.3 FGFR3 FGFR3 Includes previous San Diego type Thanatophoric dysplasia type 2 (TD2) AD 187601 4p16.3 FGFR3 FGFR3 Severe achondroplasia with developmental delay and acanthosis nigricans (SADDAN) AD See 187600 4p16.3 FGFR3 FGFR3 Achondroplasia AD 100800 4p16.3 FGFR3 FGFR3 Hypochondroplasia AD 146000 4p16.3 FGFR3 FGFR3 Camptodactyly, tall stature, and hearing loss syndrome (CATSHL) AD 187600 4p16.3 FGFR3 FGFR3 Inactivating mutation Hypochondroplasia-like dysplasia(s) AD, SP Similar to hypochondroplasia but unlinked to FGFR3, probably heterogeneous; uncertain diagnostic criteria See also group 33 for craniosynostoses syndromes linked to FGFR3 mutations, as well as LADD syndrome in group 39 for another FGFR3-related phenotype 2. Type 2 collagen group and similar disorders Achondrogenesis type 2 (ACG2; Langer–Saldino) AD 200610 12q13.1 COL2A1 Type 2 collagen Platyspondylic dysplasia, Torrance type AD 151210 12q13.1 COL2A1 Type 2 collagen See also severe spondylodysplastic dysplasias (group 13) Hypochondrogenesis AD 200610 12q13.1 COL2A1 Type 2 collagen Spondyloepiphyseal dysplasia congenita (SEDC) AD 183900 12q13.1 COL2A1 Type 2 collagen Spondyloepimetaphyseal dysplasia (SEMD) Strudwick type AD 184250 12q13.1 COL2A1 Type 2 collagen Kniest dysplasia AD 156550 12q13.1 COL2A1 Type 2 collagen Spondyloperipheral dysplasia AD 271700 12q13.1 COL2A1 Type 2 collagen Mild SED with premature onset arthrosis AD 12q13.1 COL2A1 Type 2 collagen Often associated with p.R719C and p.G474S mutations SED with metatarsal shortening AD 609162 12q13.1 COL2A1 Type 2 collagen Often associated with the Am J Med Genet 155A (5), 943–968
18. Bone Dysplasia Ontology
19. ISDS Nosology in OWL
21. Bone Dysplasia Ontology (BDO) • augmented information contents • external links (RefSeq, Uniprot etc) • phenotypic features (HPO) • gene mutations (HGVS, NCI)
22. http://purl.bioontology.org/ontology/BDO BMC Bioinformatics. 2012 Mar 26;13:50.
23. Knowledge Evolution • domain knowledge is constantly evolving • BDO needs to be kept up to date • huge task for a single individual • solution: community curation - “crowd-sourcing”
24. The Skeletome Knowledge Base • community driven knowledge evolution platform • ‘Wikipedia for Bone Dysplasias’ • human readable front end to the BDO • allows updating the BDO on the fly • no need to publish an updated nosology in a journal!
25. http://knowledge.skeletome.org/
28. Step 2: Integrating patient data Disease Gene Patient Phenotype similar Patients
29. Skeletome Archive • patient database • specifically designed for phenotype data • based on semantic web technologies • web-accessible • encourages data sharing
34. encourage contributions through data sharing
35. other sources of patient data • existing databases • medical literature • electronic medical records
36. text mining for phenotype data - logical decomposition Biomedical Informatics Insights 2013:6 1–14 PLoS ONE 2013, 8:e55656.
37. Step 3: Integrating other knowledge sources Disease Gene Patient Phenotype similar Patients Mouse Pathway Drug
40. SPARQL Query Builder
41. Step 4: Building User Interfaces Disease Gene Patient Phenotype similar Patients Mouse Pathway Drug
42. Data Visualisation MediViz Conf Proceed 2007, 43–47.
44. Machine Learning • Decision support using ESDN data: • predict diagnosis based on clinical features • training set: 283 ESDN cases, 6 different diagnoses • multiple machine learning algorithms • one optimised for rare disease phenotype data PLoS ONE 7(11): e50614. doi:10.1371/journal.pone.0050614
45. clinicians 39.66 computer outperforms clinical geneticist PLoS ONE 7(11): e50614. doi:10.1371/journal.pone.0050614
46. Bio-LarK Platform" Biomedical Large Knowledge Collider Disease Gene Mouse Patient Pathway Drug Phenotype similar Patients Expression
 profiles Metabolites
47. BioLarK Team Tudor Groza Craig McNamara Razan Paul Andreas Zankl