These results are from the paper
A. Brady, K. Maxwell, N. Daniels and L. Cowen, Fault Tolerance in Protein Interaction Networks: Stable Bipartite Subgraphs and Redundant Pathways PLoS One 4(4): e5364. doi:10.1371/journal.pone.0005364 2009.
Please cite our paper if you use them.
We analyzed two yeast synthetic-lethality interaction networks. The first was the SL graph induced by the subset of the yeast interactome which was studied by Kelley and Ideker (Nature Biotechnology, 2005). The second was an SL graph brought up to date with all the ORFs and synthetic-lethal interaction data contained in the BioGRID release 2.0.33 (October 2007).
Browse BPMs generated by our method -- and examine functional enrichment results -- here (for both networks).
Look here for p-values characterizing the distribution of new SL interactions across the BPMs we generated for the Kelley-Ideker graph. Given a BPM X = (setOne + setTwo) generated from the KI interaction graph, and given the set newSL of synthetic-lethality interactions which have been discovered between all possible pairs of nodes in X, let newSL1 be those interactions in newSL which are between nodes in setOne; let newSL2 be the interactions in newSL which are between nodes in setTwo, and let newSL1-2 contain interactions with one node in setOne and the other in setTwo. The p-values given here represent the probability, when looking at a random graph on |setOne| + |setTwo| nodes and |newSL1| + |newSL1-2| + |newSL2| edges, of seeing as many or more SL interactions in newSL1-2 than were observed. Note that the distribution of the SL interactions in newSL is completely independent of the construction of X, which was built without knowledge of these edges, so low p-values observed here (indicating a tendency of SL edges to cross between pathways as opposed to remaining within them) provide strong evidence supporting the hypothesis that the associated BPMs have a consistent biological meaning.
Look here for p-values characterizing the distribution of physical interactions across the BPMs generated for both networks. Here, p-values quantify the likelihood of seeing as many or fewer physical interactions crossing between setOne and setTwo as were observed. Note that while these interactions were used to construct candidate seeds for BPMs in both the Kelley/Ideker and Ulitsky/Shamir papers, they were totally ignored by our method in constructing BPMs. Low p-values here (indicating a tendency of physical interactions to remain within pathways rather than crossing between them) thus provide even more independent evidence supporting the idea that the associated BPMs are biologically meaningful.
Look here for p-values characterizing the
distribution of synthetic-sick interactions across the BPMs generated
for the new network. Here, p-values quantify the likelihood of seeing as
many or more synthetic-sick interactions crossing between
setOne and setTwo as were observed. Note that while
these interactions were used to construct and/or score BPMs in both the
Kelley/Ideker and Ulitsky/Shamir papers, they were totally ignored by
our method in constructing BPMs. Low p-values here (indicating a
tendency of synthetic-sick interactions to remain between
pathways rather than remaining within them) again provide more
independent evidence supporting the idea that the associated BPMs
are biologically meaningful. Note, however, that in a handful interesting
cases where there are enough interactions for statistically significant
conclusions to be drawn, the synthetic-sick interactions do not
respect the structure of the putative BPM containing them.