Ribosomal RNA contamination

[tashia]

Hi everyone,

I have a question in relation to a differential expression analysis that I ran with rRNA contaminated samples.

Essentially, my samples had 60-70% rRNA content, but I chose to go ahead with the analysis once removing these sequences since I had enough reads to work with (15-20M). Contamination was seen in both cases and controls.

However, what I am wondering now is if I can trust the results I am seeing? Is there any way that high rRNA content can affect patterns of expression? Comparing results with cases and controls with no contamination I see 1. More DEGs in the non-contaminated sample 2. Different DEGs (only 1 gene overlap). Of course they are different people but they do have the same disease and the controls are matched so I am a bit at a loss to whether this makes sense

Hope this was clear.
Many thanks.

[tashia]

Bumping my own post.

[ddb]

Can you give some more details about your experiment set up and the methods / pipeline you are using. Is it possible that in your contaminated samples you do not have enough coverage to identify most of the differentially expressed genes?

[tashia]

Can you give some more details about your experiment set up and the methods / pipeline you are using. Is it possible that in your contaminated samples you do not have enough coverage to identify most of the differentially expressed genes?

I should have enough coverage in the contaminated samples. I had approximately 75M reads before removing the rRNA sequences.

The pipeline is removing rRNA sequences with bbduk, trimming with trimmomatic, aligning with STAR (latest HG38 version) counting with HTseq and finally DESeq2. About half of the samples were contaminated but were sequenced at a greater depth to achieve the same coverage as in the non-contaminated samples once removing the rRNA content. A PCA plot shows that the contaminated samples are clearly distinct from the "good" batch. I have controlled for batch (and tried several tools) but the PCA plot still shows a clear distinction. It feels like either this seperation is due to that these samples were sequenced at two different times (but many people work with samples like this and I think I should be able to control for this) OR due to the fact that they had a lot of rRNA during sequencing.

Do you think that expression levels can be affected by a high content of rRNA in the sample while sequencing?

[GenoMax]

Do you think that expression levels can be affected by a high content of rRNA in the sample while sequencing?

Not likely. For a sequencer it is just DNA. Does not matter if it came from rRNA or some other gene. If the two sets of runs happened on two different chemistries or different machines then may be something else to consider.

If your samples were all treated the same how come you have rRNA contamination in some samples but not others?

[tashia]

Not likely. For a sequencer it is just DNA. Does not matter if it came from rRNA or some other gene. If the two sets of runs happened on two different chemistries or different machines then may be something else to consider.

If your samples were all treated the same how come you have rRNA contamination in some samples but not others?

Thanks for your reply.
Unfortunately the rRNA depletion failed in about half of the samples and instead of re-doing the libraries the sequencing facility sequenced at a greater coverage instead...

They should have been run on the same machine and the only difference is that they were run at different times and then of course the contamination. However, if the rRNA contamination shouldn't cause any problems, I don't know why I can't control for this batch effect that I see in the PCA plot I have tried including batch in the model in DESeq2 and EdgeR. I have also tried correcting with RUVseq and ComBat. I am now looking into ImpulseDE2.