Dear all,

about correcting the batch effects in LIMMA and SVA, 'd appreciate having your comments :

assuming that we have a set of RNA-seq data (no treatment, + treatment) in many distinct BATCHES, would the linear model in LIMMA :

design <- model.matrix(~0 + BATCH + Treatment, data=RNA)

suffice to correct for batch effects, or shall we additionally use COMBAT (as it is described in the SVA package : https://bioconductor.org/packages/release/bioc/vignettes/sva/inst/doc/sva.pdf ) in order to obtain the corrected counts ?

thanks a lot,

-- bogdan

I would myself just put BATCH in the limma linear model and omit Combat, and that is also recommended by Vegard Nygaard's group:

https://support.bioconductor.org/p/72926/

The Combat author (Dr Johnson) has agreed with this advice unless there are large difference in batch variances, in which case he has recommended using Combat but also including BATCH in the limma linear model:

https://support.bioconductor.org/p/72815/

I'm not sure how that would be done for RNA-seq though. I guess Combat would be run on logCPM values, in which case you would have to follow it up with limma-trend rather than limma-voom.