[BioC] LIMMA: Suitable measure of error from result object. stdev.unscaled?

[Gordon K Smyth]

Dear Dan,

stdev.unscaled needs to be scaled by the residual standard error for each 
gene (hence the name), so to get standard errors for the coefficients you 
need:

   se.coef <- sqrt(fit$s2.post) * fit$stdev.unscaled

The moderated t-statistics are just  fit$coef / se.coef.

Best wishes
Gordon


> Date: Thu, 13 May 2010 16:28:36 +0100
> From: Daniel Brewer <daniel.brewer at icr.ac.uk>
> To: Bioconductor mailing list <bioconductor at stat.math.ethz.ch>
> Subject: [BioC] LIMMA: Suitable measure of error from result object.
> 	stdev.unscaled?
>
> Hello,
>
> I have a 2-colour microarray experiment with a complex design.  I would
> like to visualise the estimated coefficients and associated error with
> the most significant genes that come out as result of LIMMA (lmfit,
> contrasts.fit, eBayes).  I thought that it should be stdev.unscaled, but
> this seems to be the same for all the genes, which I don't think makes
> much sense.  What is an appropriate way to calculate the estimated error
> associated with a coefficient?
>
> Many thanks
>
> Dan
>
> -- 
> **************************************************************
> Daniel Brewer, Ph.D.
>
> Institute of Cancer Research
> Molecular Carcinogenesis
> Email: daniel.brewer at icr.ac.uk
> **************************************************************

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