[BioC] Implications of design argument to limma::voom?

[Gordon K Smyth]

Hi Ryan,

The design matrix to voom must be the same as you use for lmFit (or be an 
alternative parametrization of the same model), otherwise voom is 
analysing the wrong variances.

Best wishes
Gordon

------------------ original message --------------------
[BioC] Implications of design argument to limma::voom?
Ryan C. Thompson rct at thompsonclan.org
Wed Jun 5 01:03:27 CEST 2013

Hi,

I'm using limma's voom function to analyze some of my data, and I'm
wondering about the implications of the design argument to voom. The
default design is an intercept-only design, i.e. a model formula of ~1.
However, I can also pass in the same experimental design that I will
pass to lmFit later (or conceivably any other design). I'm not entirely
sure what assumptions I'm making by choosing either of these options.
Can someone familiar with such things explain what the implications of
passing or not passing a design to voom are?

Similarly, I have the same question for DESeq's
varianceStabilizingTransformation: I can use estimateDispersions with or
without a design matrix and then do the VST either way, and I'm not sure
what assumptions I am making in either case.

Thanks,

-Ryan Thompson

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