[BioC] edgeR, logFC calculation in factor combination

[Gordon K Smyth]

Dear Mike,

On Fri, 30 May 2014, Mike Miller wrote:

> Dear Gordon,
>
> Thank you for your answer. I still have one more question regarding output
> of the lrt function.

...

> I am having difficulties in understanding how the testing works. I find 
> it necessary to understand it in order to see which samples 
> (combinations of which factors) will be used for qPCR validation.

Yes seem to assuming that you can reproduce the results of the edgeR 
analysis using just a subset of the samples.  I am a bit puzzled why you 
would think that.  You will need all the samples to reproduce the same 
results.

> I'll try to be precise in describing what still troubles me: Assuming 
> that there are 2 blocks: block_1 and block_2, and the "final list of DE 
> genes" is actually the list of genes (the output of lrt()) with a 
> certain p-val assigned to that gene after testing, and assuming that the 
> design is as previously described:
>
> design_combi=model.matrix(~0+ ctrl_loc + Gender+Sample_blocking,
> data=combi),
>
> my question would be how this final list of DE genes is produced: are 
> those all the genes that are differentially expressed (DE) between 
> A_disease and B_disease if the comparison takes into account only males 
> in the block_1 plus all the genes that are DE between A_disease and 
> B_disease if the comparison takes into account only males in the 
> block_2, plus all the genes that are DE between A_disease and B_disease 
> if the comparison takes into account females in the block_1 plus all the 
> genes that are DE between A_disease and B_disease if the comparison 
> takes into account females in the block_2?

No, it's nothing like that at all.  The model that you have fitted assumes 
that the same genes are DE in all blocks and all genders.  Therefore the 
software accumulates information from all the blocks and all the samples.

edgeR fits a log-linear glm to the data -- the mathematics behind this is 
explained in the published papers.

Best wishes
Gordon

> Thanks a lot in advance!
>
> Best,
> Mike

______________________________________________________________________
The information in this email is confidential and intend...{{dropped:4}}