Gene Ontologydriven similarity for gene expression correlation analysis

1
Gene Ontology-driven similarityfor gene
expression correlation analysis
Francisco Azuaje, Haiying Wang, Olivier
Bodenreider and Joaquín Dopazo
2
Two kinds of gene information

• Gene expression data

• Annotations

3
Gene expression analysis

• Clustering of genes based on co-regulation
• Similar expression patterns represent
• Common pathway
• Common response to experimental conditions

4
Annotation analysis
5
Limitations of annotation analysis
Not reflectedin the analysis
6
Semantic similarity in taxonomies
7
Approaches to computing semantic
similarity/distance in taxonomies

• Edge counting
• Intuitive
• Requires density to behomogeneous in the
  taxonomy
• Information-theoretic metrics
• Grounded in information theory
• Compensates in heterogeneity in the taxonomy

d1
d2
d1
8
Information-theoretic approaches

• Information content (IC) nodes high in the
  hierarchy have a small IC
• The information shared by two nodes can be
  represented by their common ancestors(least
  common subsumer)
• The more information two terms share, the more
  similar they are

9
Computing information content

• Taxonomy
• Frequency distribution of the nodes in a
  corpus/database
• Information content of C based on the
  probability of finding a descendant of C in the
  corpus/database

10
Information content in GO

• Taxonomyhierarchy (DAG) of is a part of
  relations
• Frequency distribution of GO terms annotation
  databases

11
Semantic similarity in GO
Lord et al., PSB 2003 Wang et al., CIBCB 2004

• Based on the information content of the least
  common subsumer (LCS)
• Several variants
• Resnik (1995)
• Lin (1998)
• Jiang Conrath (1997)

12
Semantic similarity among gene products
sim(c,c) sim(c,e) sim(d,c) sim(d,e)
annotations
g2
g1
SIM(g1,g2)
13
Experiment
14
Comparing gene-gene similarity

• Gene expression data(similar expression levels)

• Annotations(high semantic similarity)

15
Experiment

• Hypothesis Pairs of genes exhibiting similar
  expression levels also tend to have high semantic
  similarity
• Dataset from Eisen et al., 1998
• expression responses to several perturbations
  inS. cerevisiae (2460 gene products with GO
  annotations)
• Method for each pair of genes (gi, gj), compute
• absolute correlation value between gi and gj
• semantic similarity between gi and gj
• (on each GO hierarchy and for each metric
  separately)

16
(No Transcript)
17
(No Transcript)
18
(No Transcript)
19
Summary of results

• Confirmation of our hypothesis
• High similarity values are significantly
  associated with strong expression correlation
  values
• Weak similarity are significantly associated
  withlow expression correlation values
• Additionally
• Similar results were obtained
• For different number of intervals
• With the three metrics tested
• For the three GO hierarchies
• This trend is significantly stronger in the case
  of the highest expression correlation values

20
(No Transcript)