Building and Analyzing GenomeWide Gene Disruption Networks

1
Building and Analyzing Genome-Wide Gene
Disruption Networks

• J. Rung, T. Schlitt, et al. (2002)
• Presented by Sean Whalen, 2/26/03

2
Outline

• What is a gene network
• What is a disruption network
• Building the network
• Observations
• Degree distribution
• Connectivity
• Review
• Conclusions

3
What is a gene network?

• Directed Acyclic Graph (DAG)
• Nodes/VerticesObjects, Edges/ArcsRelationships
• Arbitrary meaning is assigned, in order to
  visualize relationships in a system (and acquire
  knowledge)
• Gene networks simply model genetic relationships

4
More on Gene Networks

• How to represent the network? Arbitrary.
• Example Edge between nodes means parent codes
  for transcription factor
• Example Edge between nodes means change in
  expression level of parent affects level of child
• Different modeling methods
• Bayesian, Dynamic Bayesian
• Problem only deals with small data sets
• This papers method simple, genome-wide
  analysis, demonstrated biologically meaningful
  (yeast)

5
What is a disruption network?

• Gene network built from expression data (mutant
  strain vs. control)
• Nodes are genes, edges indicated a causal change
  in expression level
• Represented as a matrix
• A discretized matrix is built from this matrix,
  to infer connectivity properties
• Disruption networkgraph representation of
  discretized matrix

6
Building the Network

• Expression data matrix
• rij log( lij / cij )
• rij jth element of ith row
• l exp. level in mutant
• c exp. level in control
• Discretized matrix
• Expression level up, down, or unchanged
• Normalize rij, adjust for gene-specific standard
  deviation
• Select cutoff level ? 2..4

• Expression matrix ? Normalize ? Select Cutoff ?
  Discrete Matrix

7
Building the Network (cont.)

• Disruption network ?' is representation of
  discretized network as a graph
• Edge between gi and gj if dij ? 0
• Label edge as down regulating if dij-1, up
  regulating if dij1. Nodes labeled w/gene names
• Expression data from all genes in a yeast mutant
  (single gene deletion) taken over 300 experiments
  w/63 control experiments

8
Matrix ? Graph Example
9
Observations

• High out degree influence many other genes
• High in degree complex regulation
• Distribution of total degree follows power law
  (scale-free topology)
• 50 of genes show change in expression with
  single deletion
• Few genes with high in AND out degree
• Strongy connected subnets (hubs) are
  evolutionally more conserved

10
Degree Distribution
11
Out Degree vs. In Degree
The point? Rare for node to have high ranked in
degree AND out degree. Only 1 nodes in degree
is in the top 50 of in degrees, AND out degree
is in top 50 of out degrees.
12
Connectivity

• How connected is the graph with different ?
  values?
• ?lt3, one big component
• Remove top 1, 5, and 10 of highest degree
  genes
• For 3lt?lt3.6, biggest component still order of
  magnitude higher

13
Sample hub (?4, rdown, gup)
14
Review

• A disruption networks is a graph representation
  of a discretized expression matrix, with a degree
  cutoff ?
• Allows genome-wide analysis
• Power-law distribution of edges
• High out degreegene encodes regulatory proteins
• High in degreegene involved in metabolism

15
Conclusions

• Disruption networks suggest scale free topology
  in gene regulatory networks
• Dominated by single large component (hub)
• Looking for subnets containing genes involved in
  a process allowed prediction of genes with
  similar functions
• DNs offer a different perspective of expression
  data than tradition methods such as heirarchical
  clustering