trimming unreliable ends of reads

[Kotoro]

What methods are normally used to trim off the unreliable poor-scoring ends of reads? Is there a tool that can statistically assess read scores and make this decision on a per-read basis, or is a cut position globally decided on?

Are there any special considerations for paired-end reads? (they've already been split and de-convoluted by barcode by the pipeline in our university's sequencing core.)

[tonybolger]

What methods are normally used to trim off the unreliable poor-scoring ends of reads? Is there a tool that can statistically assess read scores and make this decision on a per-read basis, or is a cut position globally decided on?

Something like a sliding window is good - so if you get a single bad cycle you don't cut a high quality region, but persistent rubbish is trimmed.

Generally you want to do it on a per-read basis - 'local' factors often influence a particular read, and there's no advantage to 'one size fits all'.

Are there any special considerations for paired-end reads? (they've already been split and de-convoluted by barcode by the pipeline in our university's sequencing core.)

Normally you might want to handle 'unpaired' reads separately - reads which survive QC but their partners didn't.

Blatant ad: if you're working with illumina data, i've released a tool, Trimmomatic, found here, which does what you need

[francois.sabot]

Hi
Have a look at the last version of Cutadapt, which uses a nice system for trimming, and cut only if the fater bases are of better quality than the previous... See their explanation : http://code.google.com/p/cutadapt/

[gaffa]

A third alternative is SolexaQA (http://solexaqa.sourceforge.net/), which can trim either down to "to the longest contiguous read segment for which the quality score at each base is greater than a user-supplied quality cutoff" or using the BWA trimming algorithm (which also can be performed by BWA in conjunction with read mapping).

[mmartin]

... or using the BWA trimming algorithm ...

I'd like to point out that the quality trimming in cutadapt is simply a reimplementation of BWA's algorithm. You could use the quality-trimming part of cutadapt without trimming adapters by providing an adapter sequence that's certain to not occur -- just use -a XXXXXXXX or something like this (these are literal "X" characters).

See also file lib/cutadapt/qualtrim.py in the cutadapt distribution which also shows the algorithm and contains an explanation of it.