Inferring Cellular Processes from Coexpressing Genes

1
Inferring Cellular Processes from Coexpressing
Genes

• Daniel Korenblum
• November 26, 2001

2
Motivation for Clustering

• High throughput experiments
• Reduce complexity by coarse graining
• Extract essential features
• Visualize data matrix entries with efficient
  display
• Obtain similarities that reflect biological
  properties

3
1998 Eisen, Spellman, Brown, Botstein

• Average Linkage Clustering of Time Courses
• Correlation measures similarity (scale invariant)
• Fixed offset
• Genes assumed symmetric with respect to changes
  from reference state
• Reorder genes
• Permute rows of expression data matrix
• Proximity corresponds to similarity

4
What determines the Patterns

• Assess the significance of the clusters
• Could results be statistical artifacts?
• Swap matrix elements
• Apply clustering algorithm
• See different patterns
• No prolonged correlations
• Signal from different conditions counteracts
  noise from single observations and cDNA
  variations
• Biologically interpretable implies significant

5
Gene Shaving

• Avoids a single reordering for all genes
• Different genes may require different measures of
  similarity
• Use the principle component of a set of genes
  (eigengene) as a reference state
• Select genes with high covariance with the
  eigengene

6
Gene Shaving, Cont'd

• High variation across samples
• Strong correlation across genes (coherence)
• Hierarchical methods address variations over
  samples
• Supervising affects average gene effects to
  select strong contributions on predictvie
  abilities

7
Conclusions

• Change in methodology over the past few years
• Array data holds comprehensive picture of
  cellular processes