How to compare coexpression networks from two conditions with different sample sizes

Question

[Thind amarinder]

I have gene expression data from two conditions having different sample sizes i.e 8 and 32. I want to build a correlation among all genes (for both conditions) for comparison. So, I used "psych" package from "R" which use t-test based statistics for p-adjusted values calculations.

Is this approach correct for the comparison of the networks (after taking significant p-adjust) between two experiments of different sample sizes?

or should I use another approach for p-adjusted calculations? If yes, what would you suggest?

[Kevin Blighe]

Ciao / buonasera, what are you testing with your psyche t-test? I would create the correlation networks separately, and then compare (between the network for each condition) metrics such as vertex degree, hub score, community structure, etc. A simple pipeline, here: Network plot from expression data in R using igraph

[Thind amarinder]

Ciao I am not exactly comparing networks. but want to compare interactions of particular set of genes (of my interest) with others (if any), above certain threshold of correlation . whether interacting genes are mostly the same or different in both condition. since sample size is different between conditions, so can I directly compare or not? if, no How I can justify this comparison.

[Kevin Blighe]

You want to correlate certain genes in one condition with those in the other condition?

[Thind amarinder]

Yes, how can I go about this?

[Kevin Blighe]

You can correlate together where the dimensions are the same, i.e., on the genes. So, probably:

cor(t(x), t(y))

My preferred way would still be to produce a correlation network using all genes and separately for each condition, and then compare your genes based on vertex degree, hub score, and community structure.

[Thind amarinder]

How do you calculate significance ( p-adjusted values) for correlation?

[Kevin Blighe]

There is a cor.test() function, which functions in the same way as the cor() function. After you derive your p-values, you can then adjust for false discovery with p.adjust()

[Thind amarinder]

Thank you, it seems the way "psych" calculated p.adjust is similar.

[adR]

@kevin Blighe, could please give some more information on these vertex degree, hub score, and community structure? I am just doing correlation for two different conductions with the same number and type of genes and would like to understand how they can help me to compare my correlation matrixes Thank you!

[Kevin Blighe]

Hi, have you searched the igraph documentation?