[igraph] Evaluating the functional significance of greedy NG communities in a network

[Giuseppe G.]

Hi,

I realise this question does not strictly relate to the scope of this 
ML, however I was still hoping somebody could help me with this.

I have used greedy modularity maximisation to obtain a partition of 
communities for my protein network.

Separately, I have used a semantic based approach to detect the 
functional "relatedness" of each pair of proteins in the network. So for 
each pair of interactors in the net, I have a value from 0 to 1 telling 
me how close those two nodes are in terms of some metric of the 
annotation in Gene Ontology.

Now I would like to evaluate the significance of my communities from a 
functional point of view, using these scores.

Would it make sense to do the following: for each community, get all the 
 functional scores for of all its interacting pairs. Then get the same 
number of interacting pairs observed in the community, but at random 
from the network. The run a Mann–Whitney U between the two vectors of 
scores.

Another option would maybe be to randomise x1000 times the original 
network (degree conserving), rerun the community finding and compare the 
distribution of some functional value for the random networks to the 
original net, and then get  z-score. The problem is that I don't have a 
"summary" metric to compare to some null network models: I only have 
similarity scores for pairs of interactors.

Any suggestions would be greatly appreciated.

Thanks
G

--
The University of Edinburgh is a charitable body, registered in
Scotland, with registration number SC005336.