Lecture 9

• Title: what to do with the networks!?!

MCODE

1. Vertex weighting: weight is high if has high core
2. module detection - specify threshold (e.g. 2), decision which modules are included
3. post processing - filter out, look at density, modules may overlap; haircut, remove vertices that are singly connection

other algorithms

• MCL
• k-means - divide into k groups
• pinnacle-z

function prediction

• predict function by looking at the neighbors
• Biological roles
  • GO - divide processes, cellular component, molecular function - Ashburner, nature genetics 2000
  • KEGG - manually curated biological processes
  • MIPS

function-prediction methods

• majority-rule - consider neighbors and take most frequent annotation.
• neighborhood enrichment - look at all proteins within a particular radius
• minimum-multiway cut - find multicut minimizing crossing edges
  • helps decide which nodes belong to which modules
• functional flow - choose start module and push water through the network, helps decide
  • this one looks good!

networks and gene expression

• party hub - multiple proteins at once - high correlation of expression with neighbors
• date hub - one interaction at a time - lower correlation, most likely a regulator
• if presence highly correlated, may not interact but may be part of similar pathway

networks and disease

• scale-free networks, most nodes low degree, a few hubs, very robust
• disease perturbed subnetworks

SECTION 2: CYTOSCAPE

• MCODE plug in to cytoscape to help figure out subnetworks/modules
• BiNGO - enrich networks with gene ontology