Hi!

I want to use the RUVSeq package to remove the unwanted variation in my RNA-seq data. Should I apply the function RUVg on raw counts or normalised counts (UQ normalised or TMM normalised)?

Similarly, I want to use the RUV2 function in the ruv package to remove the unwanted variation in my micro-array data - should I apply the function on unnormalised data or normalised data?

Cheers

Paul

Hi Paul,

in the RUVSeq vignette we do scale for UQ prior to compute the RUVg factors. But on our data, RUVg without scaling leads to similar results. The only thing to keep in mind is that if you do scale before computing the RUV factors, you then need to remember to put an offset in the model via, e.g., the calcNormFactors function of edgeR. This is detailed in the vignette. If you decide not to scale, then in principle the RUV factors will account for sequencing depth too, so you may want to force edgeR's offset to 0.

As for RUV-2, I believe the results are similarly robust to prior normalization. I would apply RUV-2 on the normalized data, but I think you could get reasonable results on the raw data, as well.

Best,

davide