19 months ago by
Yes, you can use countsFromAbundance="lengthScaledTPM" or "scaledTPM" as input to DESeqDataSetFromMatrix.
I prefer the DESeqDataSetFromTximport approach myself (which calculates an offset matrix for the user, and uses the estimated counts directly), but it's not a big difference. The counts + offset approach is the typical way to deal with biases in a GLM setting.
The countsFromAbundance approach instead scales the counts up and down per gene to account for any change in gene length due to differential isoform usage. Because this event is not that common or severe -- globally -- it doesn't end up scaling the counts that much for most genes. But you should still use one or the other approach (counts + offset or countsFromAbundance) to protect against the potential bias if there were differential isoform usage among isoforms of substantially different length across samples for a particular gene (e.g. as described by Trapnell 2013).