I have an experiment where I want to compare expression levels between homologous genes within the same sample. This has been done before using DESeq2 (http://doi.wiley.com/10.1111/nph.14836). The idea is to make a count matrix where each homologous gene pair is placed side by side on the same row, and then use an offsets matrix incorporating both gene lengths and library sizes.
E.g. if I had 4 samples and 1000 gene pairs then the count matrix would have dimensions 1000 x 8, the design is then ~ 1 + homolog + sampleId
(where levels(homolog)
is c(0, 1)
). This approach seems to work well with DESeq2.
However, since I have many samples DESeq2 is becoming too slow, so I wanted to try the (in my experience faster) limma-voom workflow instead. But I do not find any way to incorporate gene lengths into the limma-voom analysis. Gene lengths of homologous gene pairs often differ slightly, and not taking this into account leads to many false positives.
Is there any way to input offsets into limma-voom? Or to somehow account for both sequencing depth and gene length differences?
If my question is not clear, feel free to ask for clarification.
Thanks,
Sam