[BioC] multiple level factors contrast in nbinomLRT, DESeq2.

[Simon Anders]

Hi

On 20/08/14 12:43, sh. chunxuan wrote:

> I am not sure the correct way to interpret the nbinomLRT results in
> multiple level factors condition.

Is there a reason why you decided to use 'nbinomLRT' rather than
'nbinomWaldTest', which is suggested in the vignette for standard use cases?

> Here is a toy example. I would to find DE genes after controlling
> batch effect in the experiments, in which there are multiple levels.

In your case, the LRT allows you to find genes which are affected by the
experimental condition in _some_ way, i.e., for which you can reject the
null hypothesis "This gene is expressed at the same level in _all_ six
conditions; i.e., it is not affected by the experimental condition at all."

However, you want to perform specific contrasts:


> 1.) I would like to get the DE genes between B and A, while
 > controlling for the batch effect. Should I take "res.1" or "res.2", or
 > both are wrong? and what is the correct way to do it?

This is much easier with a Wald test. You just run 'nbinomWaldTest' and
then ask for

results( dds, contrast = c( "condition", "B", "A" ) )


   Simon