10 weeks ago by
Weill Cornell Medicine
in the case of a global shift of expression the preliminary scaling step that we show in Section 2.3 should be avoided, as it assumes that the majority of the genes are not DE.
In practice, it means to avoid the
betweenLaneNormalization and the
calcNormFactor steps. You may want to set the all the size factors to 1 in
edgeR so that no offset is added (the factors inferred by RUV should be enough).
As for your second question: yes, the
normCounts() function returns normalized expression levels that are comparable across samples in the same dataset (but not across genes).