Adaptor removal, trimming, vs masking

[ramirob]

I had a question about best practices regarding adaptor removing, trimming the ends of reads, and also removing short reads. All of these questions pertain preparing sequences for alignment with BWA.

1) Adaptor removal: we can either mask the adapter with Casava (puts N's) or clip it out with cutadapt, what is the best practice (masking vs. removing) for BWA?
2) Trimming at the ends: I have heard that this is important, trim a few bases at the beginning and also at the end (say 10 and 10). We can also just mask it with Casava, or trim it with some program. What is a standard program for trimming ends of reads?
3) Removal of short reads: when clipping out the adapter, sometimes this results in a few very small reads (ligated adapters perhaps), those small reads seem to give problems to BWA, we wrote a program for removing small reads but it is slow, is there any software out there that does it?

Thanks in advance,
Ramiro

[MeganS]

1) I think removing the adapters is probably better. Check the BWA documentation, but I think it treats Ns as a mismatch to the reference.
2) I trim 6 bases off the 5' end of RNA-seq (due to a concern over high error rates due to hexamer priming). For genome sequencing, I only trim the ends if the quality is low.
3) I use Trimmomatic to remove reads that are too short (and remove adapters and trim low quality). I think FastX-toolkit has something as well. Whatever you use, make sure it handles the orphaned reads. BWA determines read pairs by the location in the file, not by the sequence identifier.

[thomasvangurp]

Hi MeganS,

Do you have any references that indicate the source of this hexamer mispriming?

Cheers,
Thomas

[MeganS]

Here are a couple posts/blogs with some references included:

http://seqanswers.com/forums/showthread.php?t=11843

http://www.genomesunzipped.org/2012/03/questioning-the-evidence-for-non-canonical-rna-editing-in-humans.php

After reading what I could find on the topic, I am unsure if trimming the 5' end of RNA-seq is necessary, but I decided there was enough of a concern that when de novo assembling I trim 6 bases off the 5' end (a case could be made for as many as 15 bases). I do not trim for alignment to a reference.