Detecting complexes from PPI

CREATED: 200912041114 Start: 1118 End: 1137

Markov clustering simulated random walks and finds relatively dense regions

MCL for PPI found to be robust to noise and finds meaningful clusters.

Limitations of MCL:

• clusters are disjoint
• includes many noisy proteins
• does not incorporate biological knowledge

** MCL-CA #Apply MCL (parameters?) #Determine core proteins #Filter off noisy clusters (without core proteins) #Determine attachment proteins #Determine module proteins #Determine complexes (core + attachment - noisy proteins) #Rank predicted complexes (density + in-out interaction ratio)

** Result

• Two datasets from Gavin and Krogan
• Bader scoring = $|P \cap Q| / |P||Q|$
• Comparison with MCL alone
• Comparison with CORE (Leung 2009)
• Comparison with COACH (Wu Min 2009)
• Specific complexes predicted by MCL-CA ** identification of existing complexes ** novel complexes (match result from Gavin)