[Bioc-devel] Making hypothesis testing easier with design matrices?

Gordon K Smyth smyth at wehi.EDU.AU
Tue Dec 11 08:06:43 CET 2012

On Mon, 10 Dec 2012, Ryan C. Thompson wrote:

> What about this more limited proposal?

I don't see a proposal below, only a question.

> Suppose one is studying an additive model, but only one factor (or 
> equivalently, one set of interacting factors) is of interest and the 
> rest are blocking factors. For example, suppose the model is "~condition 
> + donor", but donor is just a blocking factor and only condition is of 
> interest. If one used the no-intercept formula "~0+condition+donor" and 
> set "donor" to use sum-to-zero contrasts, then am I correct in thinking 
> that the coefficients corresponding to levels of "condition" would then 
> be usable as estimates of the average logCPM for each condition?

Yes, you could do that.  But it's your business.  The power of limma and 
edgeR is that they permit you to use to fit any model and use any 
parametrization convenient to your purposes.  The results are equally 
correct regardless of what parametrization you use.

> If so, would these estimates be any better than simply computing logCPM 
> individually for each sample and taking the mean of all the samples in 
> each group?

Well, yes, because the latter would be wrong in this context.  What linear 
modelling programs such as limma and edgeR do is not "simply computing 
logCPM individually for each sample and taking the mean of all the samples 
in each group".

It makes no different to limma or edgeR what parametrization is used, 
because parametrizations are just different ways to represent the same 
underlying model.  What limma and edgeR do is to fit the underlying model.

I think that you might be better for you to try to understand what linear 
modelling is doing in a more sophisticated and complete manner before 
trying to redesign the process.  What about reading a textbook on linear 

Best wishes

> Sincerely,
> -Ryan

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

More information about the Bioc-devel mailing list