Aligning the Gene Ontologies

1
Aligning the Gene Ontologies

• Antonio Sanfilippo
• Christian Posse
• Banu Gopalan
• Pacific Northwest National Laboratory
• Richland, WA

2
Problem

• The Gene Ontologies address one knowledge domain
  at a time and are not cross-indexed
• Related concepts across distinct ontologies have
  to be explicitly used to retrieve all relevant
  information

3
Solution Develop an automatic alignment
methodology for the Gene Ontologies

• Current approached to automatic ontology
  alignment
• Based on logical structure of ontologies
• Based on the reference content of ontology nodes
• Our approach
• Establish weighted cross-ontological links using
  reference content of ontology nodes
• Calculate semantic similarity between nodes in
  the same ontology using reference content and
  ontology structure
• Align ontology nodes by combining weighted
  cross-ontological links and semantic similarity
  values

4
Establish Weighted Links across Ontologies

• Construct word-based vector signatures of GO terms

• Apply Singular Value Decomposition (SVD) so that
  vector signatures are smaller and more
  discriminative

• Derive weighted links by taking cosine measure of
  SVD vector signatures for GO terms of distinct
  ontologies

5
Calculate Semantic Similarity between GO Terms in
the Same Ontology

• Adapt entropy-based approaches developed for
  WordNet (Lord et al. 2003, Couto et al. 2003, )
• The semantic similarity between two nodes in a GO
  ontology is equal to the entropy value of their
  most immediately dominating parent (after Resnik
  95)

sem_sim(GO0007166, GO0007242)
-entropy(GO0007165) -logP(GO0007165)
6
GO Alignment

• Compute alignment values between GO nodes across
  Gene Ontologies by combining weighted links and
  semantic similarity values

BP - GO0007165 (signal transduction)
MF - GO0004871 (signal transducer activity)
GO0007166 cell surface receptor linked signal
transduction
GO0007242 intracellular signaling cascade
GO0005057 receptor signaling protein activity
GO0004872 receptor activity
sem_sim
weighted_link
alignment_strength(GO0005057, GO0007242)
weighted_link(GO0005057, GO0007242)
sem_sim(GO0007166, GO0007242)
7
A Sample Application Discover Relation between
Genes

• Prior systemic administration of
  Lipopolysaccharide (LPS) induces neuro-protection
  against subsequent stroke injury in mice (Stevens
  et al. 2004)
• LPS-induced neuro-protection involves changes in
  the expression of 12 genes
• All have similar regulation patterns
• Regulated only with LPS
• Candidates for mediators of protective phenotype
• Q How to help understand the molecular
  mechanisms of regulation for these genes?
• A One approach is to investigate ways in which
  the 12 genes are related via Molecular Functions,
  Biological Processes and Cellular Components

8
Our Goal
9
Relating Genes Via MF, BP and CC

• Data gathering
• Collect documents and GO codes associated with
  each of the 12 Genes
• GO alignment
• Using documents and GO to make alignment
  assessments
• Gene comparison by aligned GO codes
• If gene A has GO1 from BP and gene B has GO2 from
  MF, then the comparison of genes A and B will
  have GO1-GO2

10
(No Transcript)
11
Further Work

• Refine alignment approach using extraction of
  links and relations from text to inform the
  creation of GO signatures
• Carry out full alignment of the Gene Ontologies
• Evaluation
• Collect cross-ontology GO links from databases to
  establish ground truth set
• Evaluate utility of correlations between genes
  and cross-ontological links to explain the
  mechanism of LPS preconditioning
• Unify the three Gene Ontologies without loss of
  specificity in the three distinct domains
• Provide a semantic characterization of
  cross-ontological links to relate concepts across
  the three Gene Ontologies in a more meaningful
  way

12
Thank you!
13
Some Alignment Results
14
Data Set