Adapter Trimming Length Cutoff And Quality Trimming Quality Cutoff

Question

[epigene]

I want to do some quality control on raw fastq files by doing adapter trimming and quality trimming. I'm wondering what is a good read length cutoff to use to keep a trimmed reads? i.e. if after trimming, the read is discarded if its length is smaller than a cutoff.

Similar thing, what quality score cutoff should I use to trim off bad quality reads? Is 20 a good start? And is it a good idea to do quality trimming? Many aligners have quality considered so I'm not sure which option is better.

Thanks!

Answer 1

[Charles Warden]

You can also use FastQC to try and access the quality of your reads (for example, to see where the quality scores start to significantly drop off and/or at what position that drop occurs):

http://www.bioinformatics.babraham.ac.uk/projects/fastqc/

You can then carry out the length, quality, etc. trims with the fastx-toolkit:

http://hannonlab.cshl.edu/fastx_toolkit/

[epigene]

I know these tools. do you have a recommendation on the minimum length of the trimmed reads to keep?

[Charles Warden]

If you are asking about minimum final read length, you can only allow unique alignments if you want to remove ambiguous reads. Alternatively, I think the first short read sequences were ~35 bp, so I probably wouldn't go below that.

If you are asking about how much to trim off the read, I think it will depend upon your own samples.

I don't typically trim any reads when working with reference-based alignments.

For de novo assembly, it can sometimes help to trim based upon quality scores (such as a sequence of 20 or 30 nucleotides with >Q20), trim out adapter sequences, and trim out mono-nucleotide reads. However, I don't believe I have actually simply trimmed based upon length. Perhaps an extra 2-3 nt could help detect small adapter sequences, but I think there are cases where even the steps that I listed are not necessary to get good contigs (meaning that lack of any trimming might have been OK).

[epigene]

yeah, i was thinking about minimum final read length. I guess only keep unique alignments is the better option here. I think the decision to trim or not comes down to the size of the fragments. it's more important to trim for small size libraries.