This is quite dramatic.

I would look first at the logs generated during data processing. Experimental issues are usually not that clear cut, especially given the fact that this is 3'. If this is high value sample I would remap the data and redo the QC. Perhaps even someone else for help to make sure there are no issues with indexes, annotation files, etc.

But it is also very possible this is due to sample prep issues (and you are using stranded protocol). See the link.

I would also draw the graph for this individual sample "Picard" like. The heatmap is hard to interpret. It looks like there is a dramatic drop, but I would like to see exactly how dramatic it is.

I would not worry that much about ribosomal depletion, unless it is truly dramatic.

http://bitesizebio.com/13559/ngs-quality-control-in-rna-sequencing-some-free-tools/

Hi Stefan and Rohan,

Thank you very much for your answers, and thank you Stefan for the link. Yes, the heatmap is an alternative representation of the gene body coverage graphs you mentioned, I have now attached the graph here - the heatmap shows the 5' bias a little more vividly, though it can still be observed on the graph. There are two samples that show this effect most severely, however some other samples (from the same library prep batch, in which some samples suffered poor rRNA depletion) also show this effect much more mildly (though one from the same batch does not show this effect at all). I would expect to see a much lower read diversity in these samples due to the nature of the samples.

The reads were aligned using STAR. The reference was Ensembl gtf release 84 - Mus musculus (GRCm38.84). Ligation adapters were trimmed before aligning and I'm currently removing rRNA reads using computational methods so I can have another look.

Many thanks for your help,

Emma

p.s. Apologies, there are 18 samples represented on the graph which makes it a little trickier to interpret.

Wow! I have to say I have never seen such a sharp drop. I think this argues that the source is perhaps computational, though it manifests in only 2 samples. Could you do a quick bowtie+rsem on this 2 sample + 1 that looks OK? Preferably by downloading a pre-built index.

I could be wrong, but the drop is sudden I am still thinking about data processing issue.

How many transcripts you detect in each sample- can you provide a table? Just apply cutoff of 3 for reads and 0.5 for TPM.

Oh, and you can post the top 10 expressed genes, ranked by TPM along with the count and TPM data for those 2 samples along with 1 normal.

Thank you both for your suggestions! I am doing much of this for the first time as well. I will investigate further and get back to you. It's great to have had a conversation about this, so thanks again.

I'll keep you posted.

Best wishes,

Emma

I am curious if you were able to figure out what was going on?

Best,

Stefan

Apologies for the (very) delayed reply! We found that the cause of this was as follows:

The samples in question were low diversity (they were RNAs from testis cell sub-cellular compartments)

The method for sampling gene coverage only utilised 1000 genes, selected randomly from across the whole dataset. 

Only a very small proportion of genes were expressed in these samples, meaning what you see above represented a very small number of genes (single figures if I recall correctly).

Once we changed the step so that it sampled only from expressed genes in those sample, the coverage looked completely normal. This was a result of just a few genes being examined - hence the very sharp drop off.

Thanks everyone for your help at the time!

Best wishes, Emma