RSEM: identical expression for different cufflinks transcripts

Question

[Corentin]

Hi,

I ran the tuxedo pipeline on rna seq data. And I used RSEM/limma for the differential expression analysis.

I noticed something peculiar in the counts matrix from RSEM, some transcripts have exactly the same expression level for all the samples:

"TCONS_00086761"        46.23   46.23   61.40   74.93   46.23   90.77 
37.04   76.97   165.92  86.25   30.33   45.82   0.00    68.88   76.96   72.05   14.71   46.14   61.66   15.41   70.10   15.42   85.33   30.66

"TCONS_00086762"        46.23   46.23   61.40   74.93   46.23   90.77 
 37.04   76.97   165.92  86.25   30.33   45.82   0.00    68.88   76.96   72.05   14.71   46.14   61.66   15.41   70.10   15.42   85.33   30.66

When it happens it is always on consecutive transcripts (TCONS_000004 and TCONS_000005) for example, and they are related to the same gene id (XLOC). I haven't found anything about that on forums or articles.

My guess is that cufflinks misintepreted a single transcript as severals, but I am not sure. And if they were different isoforms of the same gene, they should have different expression, right ?

Does someone already noticed this and have an explanation ?

Thanks, Corentin

Answer 1

[Devon Ryan]

It's highly likely that these are overlapping isoforms (open the GTF file in IGV and have a look) where the actual alignments are almost completely in the overlapping region. In that case, RSEM will end up dividing the signal evenly over all of the transcripts. At the very least try stringTie rather than cufflinks to see if the annotation is better at these problematic spots.

[Corentin]

Thank you for your answer,

I had a look with IGV and these transcripts are indeed overlapping isoforms (they are almost the same).