Hi all,
I'm a Ph.D. in bioinformatics at Florida State University. I'm now working on some
multiple group differential expression problems using RNA-seq data of breast cancer patients. We are really interested in trying EBSeq procedure.
The
question is: when there are > 2 groups (for simplicity, let's say 3
groups "C1" "C2" "C3" as in the vignette), EBSeq handles that by first
creating all possible patterns, so here:
C1 C2 C3
Pattern1 1 1 1
Pattern2 1 1 2
Pattern3 1 2 1
Pattern4 1 2 2
Pattern5 1 2 3
and then fitting the model, and computing the posterior probability of
each gene belonging to
each pattern. But now we want to do an "
ANOVA-F type test",
that is, to find significant genes that are not EE across all
conditions. Equivalently, you may think about the likelihood ratio test
from the DESeq2 package using analysis of deviance (
ANODEV) of Negative Binomial GLM, with
the reduced model being just an intercept (~1) in the design matrix. So, I
wonder if EBSeq package has something similar to that, directly
handling significance across multiple groups instead of fitting models
for each pattern?
function, and if the fitted PP is <=0.05, that's kind of equivalent
to significance under alpha=0.05 in ANOVA F-test. But, am I really right? Also ,
I'm very concerned about how EBSeq handles multiple adjustment issue in a
Bayes context. Thanks very much for any feedback!
