Graph preprocessing

1
Graph preprocessing
2
Protein Function and Interaction Data
3
Problems with Available Interaction Data (I)

• Noise Spurious or false positive interactions
• Leads to significant fall in performance of
  protein function prediction algorithms Deng et
  al, 2003

Hart et al,2006
4
Problems with Available Interaction Data (II)

• Incompleteness Unavailability of a major
  fraction of interactomes of major organisms
• Yeast 50, Human 11
• May delay the discovery of important knowledge

Hart et al, 2006
5
Pre-processing of Protein Interaction Networks
6
Our Approach

• Transform graph G(V,E,W) into G(V,E,W)
• Tries to meet three objectives
• Addition of potentially biologically valid edges
• Removal of potentially noisy edges
• Assignment of weights to the resultant set of
  edges that indicate their reliability

Transformed PPI graph where Pi and Pj are
connected if (Pi,Pj) is a hyperclique pattern
Input PPI graph
7
Sparsification to remove spurious edges
Common neighbor- based transformation
Pruning to remove spurious edges
edges 6490
edges 95739
edges 6874
8
Related Approaches Neighborhood-based Similarity
9
References