Entering edit mode
7.4 years ago
9606
▴
330
Given a set of genes, and some hundreds of experiments in different conditions (from TCGA), I want to measure the pairwise correlation of expression of any pair of genes.
So far, I have been using the Pearson correlation, but this measure has some well known limitations. Also, it may be hard to interpret the result of anti-correlated genes.
I am planning to switch to a different metric.
Do someone have some experience with that?
Would you suggest any measure over the others??
Please give specific details. What limitations of Pearson correlation that you are worried about? Pearson measures linear correlation pretty well. Did you try some scatterplots to see what kind of behaviors emerge? Also, what do you mean by "hard to interpret the result of anti-correlated genes"? If you have already thought of some metric of correlation, then which one and why?
Can't see why you wouldn't want to use Pearsons. There is a tool WGCNA that calculates gene networks using pearsons. It depends on what you are looking for, if you want to find non linear relationships there are methods built on mutual information like ARACNE. However, normally pearsons is fine. Anti correlated relationships can be taken into account by building 'signed' networks using WGCNA I believe.
Exactly. Pearson can only capture linear relationships. Mutual information criteria could be a better choice, I was wondering if some particular criterion has been adopted in the literature.