Hi there,

Just got a couple of questions for DESeq2:

1. To build a design including paired samples (i.e. Subject) and batch variables (as estimated by SVA, i.e. SV1 and SV2), and plotting SV1 and SV2 seems to resemble batch processing on different flowcells, can the design formula be built as ~Subject + SV1 + SV2 + Condition?

2. Would the lfcShrink () coupled with lfcThreshold be comparable to the limma treat() by specifying the same lfcThreshold? therefore to enable a more direct comparison between RNA-seq (DGE using DESeq2) and microarray (DGE using limma) data for the same samples.

Thanks Guan

1) Are subjects contained within the same batch? If so, that's good and then you can just use ~subject + condition, because the sample pairing obviates the need to do batch modeling. In fact, adding SVs would be bad because these will be nearly collinear with subject and induce problems in the linear model fitting.

2) Yes, lfcShrink and lfcThreshold is conceptually similar, although we are bounding the rate of an event that the LFC is false sign or that it is smaller in absolute value than theta (we call this False Sign or Smaller, FSOS), and treat() is performing null hypothesis testing with a null region followed by a FDR bound.