I have RNA Seq data from mouse and human skin ( 2 replicates each) and want to compare the expression of the orthologous genes to find any which are differentially expressed. I have quantile normalized the gene expression matrix across all 4 samples (2 mouse + 2 human). I eventually want to calculate the log fold change in expression of all orthologous genes between the 2 species. However before I do this, I should control for gene length, right? Will this be enough to give me an idea of the differentially expressed genes or should I be employing more sophisticated methods? Any comments would be helpful. Thanks a lot.

  • $\begingroup$ This might be better suited at Cross Validated. $\endgroup$
    – MattDMo
    Aug 26, 2015 at 22:51

1 Answer 1


It depends on what type of data you have, really. There are methods developed solely for quantifying relative expression based on count data, such as using edgeR or limma-voom.

You don't need to correct for gene length to estimate fold-changes of relative expression, what you need to do is normalise by library size first (and in the process obtain log2 ((counts + 0.5)/1e+06) and then, following quantile normalisation, you can just calculate mouse - human or human - mouse to give you an estimate of fold change.

I still would recommend using something a bit more sophisticated like limma-voom for this task, though, because that also will enable you to get stuff like false-discovery rates for your fold changes.

  • $\begingroup$ I have count data. However the reason I am hesitant in using packages like edgeR or DESeq2 is because they are meant for comparing expression of same gene between 2 conditions. I on the other hand want to compare the expression of orthologous genes (diff genes with diff lengths) between 2 species. Also if I am doing quantile normalization (i.e. imposing the same distribution on all samples), do I still need to control for library size? $\endgroup$
    – I.Sethi
    Aug 28, 2015 at 15:37
  • $\begingroup$ Yes it is the norm to control for library size and then quantile normalise - you always need to control the number of counts for how many reads were sequenced per sample. Orthologs with different lengths should be fine - I mean, in any case limma-voom for instance estimates gene level differential expression without consideration of isoforms (which have different lengths). Orthologues can technically be regarded as variants of the same genes so maybe give it a go. $\endgroup$ Sep 1, 2015 at 0:34
  • $\begingroup$ Thanks for your answer. But the fact that limma ( & other count based methods) does not account for diff transcripts is a limitation (e.g: if gene A transcript 1 has 20 reads in condition A & gene A transcript 2 has 20 reads in condition B, it will not find transcript 1&2 as differentially expressed...whereas depending on the gene this can be biologically relevant.) this is why I am hesitant to use these methods for my analysis. $\endgroup$
    – I.Sethi
    Sep 2, 2015 at 13:48
  • $\begingroup$ You are going to run into similar problems anyway if you are comparing orthologues by fold changes alone - the only way to get round this is to derive exon specific counts, in which case limma has a diffsplice function to check for differential splicing. or you could use RSEM to estimate counts for comparable isoforms in your organisms and then instead of gene counts use isoform counts. $\endgroup$ Sep 3, 2015 at 14:41

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.