4
$\begingroup$

To my understanding, as long as you have overlapping paired-end reads, most modern aligners/merge algorithms can handle the alignment of each read pair well, and don't benefit from less data (i.e. quality-trimmed). In theory, more data should always be better, and as the quality drops the closer you get to the end of one read, the better it usually gets at the corresponding position on the other read. So both aligning and merging should theoretically be easier the more bases are kept, because you get more redundancy. So why is it people still quality-trim paired-end reads?

$\endgroup$

1 Answer 1

7
$\begingroup$

You're right - most aligners can handle untrimmed bases, due to partial kmer matching or local alignment. Most of the time - even if they don't do local alignment by default - partial matching is only a command-line option away.

I suspect the main reason most people continue to do trimming is because it's quite hard to get people to change their approach after they've found something that works. Quality trimming was necessary for the first versions of Burrows-Wheeler aligners, which required exact strings for matching in their default match mode, and that practise has continued.

As an aside, most aligners ignore quality scores and base their decisions on the base sequence alone. There are some exceptions (e.g. LAST in quality-adjusted alignment mode), but ignoring quality scores is the most common approach.

However, on reflection I realise that there is one thing that trimming programs do that can be useful, and that is adapter trimming, which is typically done at the same time as quality trimming.

Adapter trimming is a good idea to remove short sequence fragments that could end up as false positive matches, or disrupt genome assemblies. While a few assemblers will clip out highly-repetitive sequences as adapter-like, the process is not universally applied, and shorter adapter fragments could still be incorporated. Likewise, some mappers can also recognise common adapters (e.g. NCBI's sequence submission workflow does this as a QC step), but the process is not universally applied. There are enough different technologies that automatically doing general adapter detection in addition to other mapping can bring its own problems around false negative and false positive results.

$\endgroup$
2
  • $\begingroup$ Thank you @gringer for your answer. It seems very strange to me that most aligners would ignore quality scores, as that information should be able to help the alignment. For example I guess you could weight the importance of every position against its quality score. Or mask 3' ends with too low quality scores. (Ignoring potential technical limitations of the current underlying algorithms). But if the aligner algorithms at hand today are too limited to take quality scores into account, perhaps it's time for someone to advance the field and create a new algorithm tailored for sequence alignment. $\endgroup$
    – Joel
    Commented 16 hours ago
  • $\begingroup$ And yes, I know adapter trimming has a completely different use, that's why I specified quality-trimming here. But I agree it's still worth pointing out in this context. $\endgroup$
    – Joel
    Commented 16 hours ago

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.