[BioC] Limma and missing data

Matthew Ritchie mritchie at wehi.edu.au
Tue Mar 2 00:56:33 MET 2004


Hi Sean,

>I have a very noisy data set that I am having trouble thresholding to get
>useful genes, so I am resorting to using missing values.  (Yes, I could
>impute, but my n is not huge and consists of four non-distinct groups).  My
>question is:
>
>How does limma handle missing data?  
>
When fitting gene-wise linear models in limma using lmFit(), missing 
values are removed from the response vector of log-ratios, and the 
corresponding row/s of the design matrix are removed before the linear 
model is fitted to each gene.

>It doesn't mind NA's coming in, but
>does it correct for the different df at each gene given the number of NA's?
>
So yes, the degrees of freedom will depend on the number of missing 
values for each gene.  Best wishes,

Matt Ritchie

>Thanks,
>Sean
>



More information about the Bioconductor mailing list