Metagenomics: rationale behind Q trimming

Question

[f.a.galkin]

I am wondering on how I should trim my reads for a metagenomics study. Surely, I don't need extra stringent conditions as Q=30. But how low can I get to keep my reads numerous and long? I have considered checking average quality first and then trimming, say, at half the average quality. But then I'd have different trimming settings for different runs.

What are your ideas on balancing sensitivity/specificity in this case?

Answer 1

[finswimmer]

I am wondering on how I should trim my reads for a metagenomics study.

Don't trim your data :) As long as you don't have a significant drop-off of the quality values you would throw away informations that could be useful.

If your data is paired-end you can use bbmerge to correct quality values in regions where paires overlap.

fin swimmer

[f.a.galkin]

That makes sense I guess, I'll just set Q=4 to make sure nts are more likely to be right than wrong. I guess there's no way of finding out if it works other than testing it

[finswimmer]

Mapper/Aligner and VariantCaller have there own logic how to handle bad quality values. Just let them do their job.

In most cases you will have bad quality values at the end of your reads. If the Mapper/Aligner isn't satisfied with it, it will clipp those bases by it's own.