SOLiD paired end data and BWA - cleaning/mapping

[Kath]

Hello everyone,

I notice there has been the odd post about this general topic before but I am still confused!

I have SOLiD single and paired end data which I would like to take through the following steps:
1) Quality assessment - Fastqc
2) Data cleaning - Quality trimming/removing adapters if necessary
3) Mapping using BWA

At the moment I am converting the csfasta and qual files in to double-encoded (fakespace) fastq files for BWA. I reverse the F5 reads so BWA treats them correctly. However, can the same fastq files be used with Fastqc? Also, can CutAdapt and fastq_quality_trimmer handle this format? If anyone else out there still uses SOLiD data and can give me some details of their workflows I would be very grateful.

Best wishes,
Kath

[BrianJames]

I used to work with SOLiD data, and I have to ask why you are going to all of this trouble. In my experience, the best way to map SOLiD data is to stay in colorspace, and use Bfast, or mapreads (part of lifescope/bioscope).


Brian

[Kath]

Thanks for your reply. I have Lifescope mapped BAM files but I'm trying to dedup using Picard and am getting some very high duplication rates. I've been told that Picard seems to overestimate duplication rates in Lifescope mapped data and that BWA provides better compatibility with Picard. I did look into using Bfast but previous benchmarking studies suggested BWA would be better.

Anyway, it looks like there is a script - solid-trimmer.py - that will do what I want.

https://github.com/brentp/bio-playground/blob/master/solidstuff/solid-trimmer.py

[BrianJames]

OK. Are you working with DNA-seq, or RNA-seq? If it is DNA seq, I would be worried about not working in colorspace, if it is RNA-seq, removing duplicates from SOLiD short read data may not be a good idea. Just my experience.

Brian

[colindaven]

We tested all SOLiD mappers very thoroughly, and NovoalignCS (commercial but not too expensive) won hands-down. I would really recommend it.

BWA dropped color space support a long time ago and did not perform very well in terms of # mapped reads. Bfast wasn't much fun either.