Linking Diseases and Genes through Informatics Knowledge Bases and Ontologies

1
Linking Diseases and Genes through Informatics
Knowledge Bases and Ontologies

• Joyce A. Mitchell, Ph.D.
• National Library of Medicine
• University of Missouri

2
Research Collaborators

• Olivier Bodenreider, M.D., Ph.D.
• Alexa T. McCray, Ph.D.
• Allen C. Browne

3
Research Goals

• Investigating methods of connecting the disease
  and genomic information.
• Overall goals are to
• Overcome difficulties traversing multiple
  information resources
• Examine coverage of Unified Medical Language
  System (UMLS), Gene OntologyTM (GO),
  LocusLink-OMIM
• Develop methods to use ontologies more
  effectively
• Present data in understandable manner

4
Background UMLS

• NLM developed, maintains
• Purpose facilitate retrieval integration of
  information from multiple biomedical sources
• Interrelates 60 biomedical terminologies
• MeSH, SNOMED, Read Codes, ICD, etc
• No vocabulary focused on molecular biology
• 1.5 million English terms 800,000 concepts
• 134 Semantic Types 54 Semantic Relationships

5
Background Gene Ontology

• GO Consortium developed, maintains
• Purpose
• promoting cross-species methodologies for
  functional comparisions
• Allows annotation of molecular information on
  genes, gene products
• an essential start to creating a shared language
  of biology
• Focused on
• molecular function (5626 terms)
• biological processes (4677 terms)
• cellular components (1077 terms)
• Two semantic relations (is-a and part-of)
• Genome Research 2001 111425-33.

6
Background - LocusLink

• Curated, gene-centered resource of National
  Center for Biotechnology Information (NLM)
• Gene names, gene product names, gene product
  functions, and reference sequences (DNA, RNA,
  protein)
• Associates phenotype (diseases) to the genotype
  via Online Mendelian Inheritance in Man (OMIM)
• Online links to major bioinformatics knowledge
  bases and the literature

7
Specific Questions

• This study looked at coverage in UMLS of
• 1244 genes associated with human diseases
• 1702 diseases associated with the genes
• 11,380 Gene Ontology terms
• 38,832 genes/gene products in GO database
  (141,071 names)
• Associations of genes and their functions in UMLS
• Representation of gene function in GO compared to
  the UMLS

8
Methods

• LocusLink query
• human genes whose sequence is known and
  associated with disease (1244 loci)
• LocusLink data
• Genes/gene products (official names, synonyms,
  symbols)
• Phenotypes (diseases) (1702 diseases)
• GO data
• all concepts (ontology terms), excluding obsolete
  terms (11,380 terms)
• Gene products from all species (134,646 unique
  names, 38,832 genes)

9
Methods

• LocusLink and GO terms mapped to UMLS concepts
• normalization used
• mappings constrained by semantic type
• LocusLink loci studied for relationships in UMLS
• Gene/GP phenotype
• Gene/GP molecular function
• Gene/GP biological process
• Gene/GP cellular component
• For specific genes compared annotations in GO to
  representation in UMLS

10
Results - 1

• For 1244 genes from LocusLink
• 18 found in the UMLS

11
Results - 2

• For 1702 phenotypes (diseases) corresponding to
  1244 genes
• 34 found in the UMLS (575/1244)
• Most frequent single gene diseases covered
• Huntington Disease
• Cystic Fibrosis
• Marfan Syndrome
• Phenylketonuria
• Achondroplasia

12
Results - 3

• GO terms found in MeSH 2764 terms
• GO terms found in SNOMED 1366 terms
• GO terms found overall 27 3062/11,380
  

13
Results - 4

• For 134,646 unique gene names in GO database

14
Results - 5

• LocusLink UMLS Relationship Categories found
  overall 72

15
Results - 5

• Type of Relationship
• Associative 613
• Co-occurrence 3353
• Hierarchical 1168

16
Results - 6

• Representation of gene function in GO compared to
  the UMLS

17
Neurofibromin 2 merlin in GO
18
(No Transcript)
19
(No Transcript)
20
Discussion
21
Best Worst Mappings

• Best mapping categories
• Molecular function (GO) 44
• Cellular component (GO) 35
• Phenotype (LL) 34
• Worst mapping categories
• Gene synonym (GO) 6
• Biological process (GO) 5
• Gene symbol (GO) 2

22
Only 34 of diseases?

• In OMIM-LL, diseases are subdivided by genetic
  causes but not in UMLS
• E.g. Limb Girdle Muscular Dystrophy
• LGMD is represented in UMLS
• A SNOMED term
• in MeSH it is an entry term for muscular
  dystrophies
• MeSH notes for MD A general term for a group of
  inherited disorders which are characterized by
  progressive degeneration of skeletal muscles (ed,
  2000)

23
Limb Girdle Muscular Dystrophy genetic types
24
Only 5 of Biological Processes?

• Only 256 of the biological processes mapped to
  terms in UMLS.
• In GO, processes are elaborated organism
  specific
• Example UMLS - Mitotic spindle
• GO
• Mitotic spindle assembly
• Mitotic spindle assembly (sensu Saccharomyces)
• Mitotic spindle assembly (sensu Fungi)
• Mitotic spindle checkpoint
• Mitotic spindle elongation
• Mitotic spindle orientation
• Mitotic spindle positioning
• Mitotic spindle positioning and orientation

25
Why so few gene names and synonyms mapped?

• Official gene names have metadata and comments.
• dystrophin (muscular dystrophy, Duchenne and
  Becker types), includes DXS143, DXS164, DXS206,
  DXS230, DXS239, DXS 268, DXS269, DXS270 DXS272
• No single source has all names and synonyms
• GO synonym field contains IPI number for well
  known genes, does not match UMLS (useful cross
  reference but not a synonym)
• Symbols are short acronyms and match poorly

26
Summary 1

• UMLS needs improvement in molecular biology
  domain but has considerable content
• 27 of GO concepts map
• 34 of single gene diseases
• Existing UMLS terms come primarily from MeSH and
  SNOMED
• Overall, positive mapping for 13,000 terms

27
Summary continued

• If the terms are in UMLS, it is possible to find
  a relationship between genes and phenotypes and
  gene function much of the time.
• UMLS does better with the human genes (20) than
  with genes from all organisms (11)
• UMLS and GO representations complement each
  other.