Entering edit mode
Hi,
I am trying to compare between 10 groups each with two replicates (RNA-seq data). I wish to determine the genes which distinguish each individual apart from one another. So far looking through the edgeR manual I have not found a way to compare all of the groups at once in order to obtain values such as logFC, logCPM, PValue, and FDR. Do you have any suggestions of how to compare all of these groups to one another at once in edgeR instead of comparing in a pairwise manner?
In other words, is it possible to compare more than two groups at once using a negative binomial model?
Otherwise, do you have any suggestions of how to go about this problem?
Thanks!
On that note, an example of the coefficient approach is given in Section 3.2.6 of the user's guide.
Hi Ryan,
I used the multiple coefficient approach and it worked! I have a couple follow up questions:
1. How can you get all of the values from the lrt? It limits me to only a few of the top Tags?
2. For these top genes the P values and FDR values are zero. Is this a rounded zero or is it actually zero? Is it possible to extract the exact values out?
3. Instead of comparing the first two groups in an exact test (negative binomial model) is it possible to compare more than one group at a time as you can using the approach you outlined above?
Thanks so much for all of your help!
1. Read the help text for topTags. There are lots of options to control the output.
2. Generally there is no such thing as a zero p-value. It's probably a floating point underflow (smaller than the smallest positive number that R can handle). Also, the fact that you got such tiny p-values (smaller than 2e-300) makes me think that you probably specified the wrong coefficients or constructed your design matrix wrong, and accidentally included the intercept in your test. You should search the edgeR User's Guide for relevant examples to make sure you are performing the right test.
3. Yes, when you pass multiple coefficients or contrasts to glmQLFTest, it will do an ANOVA-like test of the combined null hypothesis that all of them are equal to zero. If you have ten groups, then you want to test the 9 coefficients or contrasts representing the differences between groups (which coefficients or contrasts you use depends on how you constructed your design matrix).