Melvin's digital garden

Mining PPI Networks

CREATED: 200701300947 Author: Ng See-Kiong

** Background

  • Genome <-> Proteome <-> Interactome (Yeast, Worm, Fly, Human)
  • Goal

To understand why/how proteins interact

*** interacting subdomains *** sequence motifs *** structural

Discover new biology

*** protein complexes *** lethal proteins *** network motifs

** Domain-Domain interaction, building block of PPI

  • Infer DDI from PPI probabilistically, given domains of proteins
  • Rosetta stone method (Nature 2000 405:824)
  • InterDom (2003)
  • Data sources: pfam, DIP, protein complexes, domain fusions (SWISS-PROT)
  • Evaluation by predicting protein-protein interactions
  • Applications ** protein interaction prediction ** intra linking in protein complexes
  • Further work ** multi-domin interaction

** PPI from sequences

  • Linear motifs (very short), structurally interacting ** Eg, proline rich motifs, leucine rich motifs
  • Approximate double clique (NP hard) by double star
  • Further work ** multi motif interactions ** domain-motif ** motif-motif from domain-domain ** cross species interacting domains/motifs

** Multi protein complexes from PPI

  • complexes are molecular aggregations that are stable
  • experimental detection is difficult
  • highly interconnected regions in the network correspond to complexes (Tong in Science 2002)
  • DECAFF (2007) ** complexes in dense and reliable PPI neighbourhoods ** dense - maximally dense (incompleteness of PPI) ** reliable - computes reliability measure (high false positives in PPI)

** Dicover lethal proteins

  • typically use graph theoretic centrality measures
  • but PPI is scale free, above measure may not work well
  • introduce functional centrality - incorporate functions of proteins

** Discover network motifs

  • biological networks are far from random ** Global: scale free, small world model ** Local: functional modules in dense neigbourhoods, recurring patterns
  • Network motifs - consequence of evolutionary mechanisms to represent a reusable functional module?
  • Motifs should be repeated (appears many times) and unique (cannot be found in random graphs)
  • Labelled network motifs, check found motifs using other annotations ** functional ** cellular localization

Links to this note