Cytoscape: Advice on best analysis methods for PPI network

Question

[Haley]

Hello there!

We are working on our first protein interaction network analyses and would like to know if there is a program/plugin available that will spatially arrange (cluster) nodes based on significant GO terms. We have found packages that arrange things in rows or color-code nodes based on GO, but so far it has always then required manual intervention to drag the nodes out into clusters and arrange shared nodes in between, etc. We want to examine some fairly large networks that will likely have several clusters, so it would be really helpful if the clustering was automated and allowed us to choose specific parent/child GO terms to cluster by from a list. Basically, we want to enrich our protein-protein interaction network with the GO terms represented in the network and, instead of looking at all of the GO terms represented, we want to find a way to ‘move up the tree’ to examine the less-specific GO terms to get a general idea of what our proteins of interest are associated with. For example, if we see there is an enrichment of GO terms related to cell division, is there an automated way to visually separate out the nodes in the network after selecting that GO term from a list? Are there any programs or Cytoscape plugins that can do this? We are new to this type of analysis so any advice would be greatly appreciated.

Thanks so much for your help!

Haley

[qmarulfiz]

Im might be wrong. I am still new in this stuff (3 month new!).

I stumble on this issue also during my play around with cytoscape in trying to learn bioinformatic stuff.

What I did was:- 1) Obtained list of genes ID. 2) Enriched GO term using DAVID. (DAVID produce a table with GO terms with its respective genes ID). 3) Using R and DAVID table, assigned ID to each GO term (I just number each GO term 1-200 if you have 200 GO term), seperate genes from the DAVID table while preserving their GO term and GO term ID. 4) Same list of genes ID, obtained PPI using StringApp. 5) Load PPI into cytoscape, Load table in no.3 into cytoscape matching gene ID. 6) Now each of your PPI nodes have GO term assigned, and GO term ID assigned. 7) I forgot what Cytoscape app I used, but I just select GO term ID I want to see, and it clustered that ID while preserving the PPI network.

Not sure this help or I understand the question correctly. Like I said, I am new here, but the question looks similar to what I am having trouble with.

I will be really interested to know on how you going to explore your results by using you network overview analysis.

Answer 1

[Jean-Karim Heriche]

There's no good way to associate a protein with only one term. The most straightforward way would be to reduce the ontology to a list of terms of interest and assign terms from this list to proteins. A potential way forward that side-steps this issue could be to compute a semantic similarity between the proteins using their GO annotations. You can then either use this alone as a graph adjacency matrix or combine it with the interaction graph (using the semantic similarities as weights of the interaction graph) depending on whether the clusters you want to identify should reflect only the functions or the combination of interactions and functions. From there you can apply a graph clustering algorithm or use semi-supervised or supervised approaches to identify groups of interest.