[BioC] edgeR glmLRT pearson residual

[Gordon K Smyth]

Dear Laura,

edgeR does not automatically provide residuals, but it does provide 
information from which you can compute them yourself fairly easily.

To get Pearson residuals:

   fit <- glmFit(y, design)
   y <- fit$counts
   mu <- fit$fitted.values
   phi <- fit$dispersion
   v <- mu*(1+phi*mu)
   resid.pearson <- (y-mu) / sqrt(v)

To get deviances residuals you will need the devel version of edgeR:

   d <- nbinomUnitDeviance(y,mu,phi)
   resid.deviance <- sign(y-mu) * sqrt(d)

Better than either of these however is the midp-quantile residual:

   resid.quantile <- zscoreNBinom(y,mu,size=1/phi)

To read about quantile residuals:

   http://www.statsci.org/smyth/pubs/residual.html

Best wishes
Gordon

> Date: Sat, 8 Feb 2014 16:53:27 -0500
> From: Laura Eierman <lee27 at cornell.edu>
> To: bioconductor at r-project.org
> Subject: [BioC] edgeR glmLRT pearson residual
>
> Hello,
>    I am interested in applying my edgeR results to a modulated modularity
> clustering algorithm to explore community structure in the significantly
> differentially expressed genes.
>
> Is there a straightforward way to calculate Pearson or deviance residual
> values from the glmLRT results?
>
> Thank you.
> Cheers, Laura

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