Taehyun Hwang and Yungil Kim

1
Revealing modularity and organization in the
yeast molecular network by integrated analysis of
highly heterogeneous genome-wide data

• Taehyun Hwang and Yungil Kim
• Csci 5980 Functional genomics and systems biology
• Apr. 21, 2008

2
Introduction

• Target problem dissection of complex biological
  systems
• Size of the underlying molecular network
• Heterogeneous nature of the control mechanisms
• Contribution Integration of heterogeneous data
  for finding modularity
• Analysis of a highly diverse collection of
  genome-wide data sets (GE, PI, growth phenotype
  data, and TF binding) ? SAMBA
• Identification of modules
• Hierarchical organization

3
Methods (1/3)

• Integrated modeling of Genomic Data
  Biclustering
• Heterogeneous data Data normalization
• Translation function continuous measurement ?
  edge probability (soft linear discretization)
• Identify significantly dense sub-graph by using
  log likelihood score modules
• SAMBA hashing and optimization
• Log-likelihood additive score ? complete
  bipartite sub-graphs
• Local search ? filtering (selecting gt80 of its
  gene-property vertex pairs)

• A weighted bipartite graph G
• Finding heavy subgraphs
• (statistically significant modules)

4
Methods (2/3)

• Functional enrichment annotation
• Functional enrichment ? GO annotations
• Each module and biological process with gt10
  annotated yeast genes
• Hypergeometric P value of functional enrichment ?
  Bonferroni correction
• Global Analysis
• Module graph nodes (all modules) and edges (gt1/3
  gene set overlap)

5
Methods (3/3)

• Example
• Biclustering using SAMBA Find modules having
  similarity of the genes only across a subset of
  the properties
• Overlapping modules ? Multiple functional genes
  analysis

6
Results(1/3) Transcriptional Network
6
7
Results(2/3) Hierarchical organization
7
8
Results(3/3) Functional Annotation
8
9
Discussion

• Does the integration of heterogeneous data always
  have synergistic effect?
• Ambiguous data integration criteria ? How?
• How can we deal with the probable contradiction
  problem among different type of data ?
• What if some of data are noisy?
• How can we decide which combination of the
  heterogeneous data are more relevant to search
  the modularity in a certain condition?
• Incorporating prior knowledge?
• Assign more weights to data based on prior
  knowledge.

9
10
Appendix for methods

• Translation function()
• The log-likelihood of a sub-graph
  ()
• The logarithm of the ratio of its probability
  under two statistical models

10