Comparison of the networks of two species

Question

[The Last Word]

Hi,

I have a set of genes from honey bees and their corresponding genes from D. melanogaster. I could construct networks using the genes in STRING. However, other than ontology, I am unsure as to what other comparisons I can do between these two networks. Please advice on some tests I could do and if possible, link me to some publications I could use. Thank you in advance.

Answer 1

[Kevin Blighe]

Between 2 networks built from the same genes, you could compare community (AKA module) structures, hub scores, closeness centrality, betweenness centrality, and vertex degree. I go over this and some other metrics, here: Network plot from expression data in R using igraph

Your question has actually come a bit premature because I am currently working on a project where I will be addressing network comparisons in a major way (tonight and tomorrow).

The famous WGCNA already has community / module comparisons implemented.

With Cytoscape Network Analyser plugin, I believe that there are yet more options.

Finally, you may want to read this: Gene co-expression analysis for functional classification and gene–disease predictions

Kevin

[The Last Word]

HI Kevin,

Thank you for your suggestions. I am wondering if biological inferences can be drawn from the information that one network is maybe smaller than the other. Maybe the centrality scores are different. That is something my supervisor always asks. Also I use Cytoscape for my analysis. I used R just to create random networks for comaprison while all the other analysis has been done on Cytoscape.

[Kevin Blighe]

What do you mean by smaller?

A major issue for network analyses is that they are mostly based on correlation (or some distance metric, like Euclidean), which are not readily conducive to the actual underlying biology. We discussed these issues in my previous lab in the US and a suggestion that came up (okay, it was my suggestion) was to somehow overlay a user's data with already well-curated pathways. This is not strictly pathway analysis, which just looks at the number of genes in each pathway and which makes no attempt to overlay (or 'impute') a user's data to a given network.

[The Last Word]

HI Kevin,

What I meant was if, say one network had smaller centrality, diameter, clustering coefficient values when compared to the other. But I guess that you have already answered that question. The problem is that I have these two networks from STRING and I was hoping for examples of network comparison work based on the different parameters that you mentioned.

[Kevin Blighe]

I see, you may consider taking a look at this pre-print in bioRxiv, where we exported a network from STRING and then further analysed it in various ways in Cytoscape:

• New insights in Tibial muscular dystrophy revealed by protein-protein interaction networks