Compare two gene co-expression networks/correlation matrizes

Question

[Benni]

Is their a way to compare two gene co-expression networks or correlation matrizes of the same genes? My idea was to just compare the pearson correlation values of the gene pairs eg. check for differences in the correlation matrix. But a pearson correlation of 0.7 could mean something different in the first data compared to the second, so maybe there is a way to normalize a correlation matrix?

[unksci]

If your comparison was about a certain number of top hits, and the values are not necessarily on the same scale, you might use the rank rather than the specific values. (Note: depending on your aim, and data, this could however also become misleading: e.g.: If you considered all pair-wise combinations, the rank of some small correlations could primarily reflect technical noise)

[Benni]

Thanks for your inputs. I found "differential network analysis", which seems to be a common tool for such a problem. Can anyone recommend this?

Answer 1

[Jean-Karim Heriche]

You can test whether the difference between the two correlation matrices is statistically significant using Jennrich's test implemented in R in the psych package in function cortest.jennrich(). Alternatively, an idea for comparing two networks would be to compare the transition matrices of the random walks on these networks.