Picard fix mate information failed with bwa mem result

[yinshe]

Hello,

I used the newest version of bwa mem (V0.7.12-r1039) to run alignment of pair-end sequencing data. The alignment is fine. However, since I added in read group information, and this was included into the command. So in the final sam file, at @PG line, it contains command line information (CL tag), it broke the SAM specification, which some tags in @RG lines show up in @PG lines because the command contains read group information. When I used picard to fix mate information, picard complained about this. Does anybody has some ways to work around this? Thanks!

[dpryan]

Can you show the relevant part of the header (ideally with a couple lines on either side)?

[yinshe]

Thanks!

The line which causes problem is:

@PG ID:bwa PN:bwa VN:0.7.12-r1039 CL:bwa mem -t 10 -R @RG ID:Sample1 CN:SequencingCenter DS:Project1 DT:2015-02-09 PL:ILLUMINA -M human_g1k_v37_decoy.fasta read1.fastq.gz read2.fastq.gz

all the items after "CL:" are separated by space, except items between "@RG" and "PL:ILLUMINA" which are separated by tab, since they are read group information.

[dpryan]

Ah, you must be running the command from within a script where you've explicitly input tabs rather than giving a string with "\t" tab designations. Try the following directly from the command line:

bwa mem -t 10 -R "@RG\tID:Sample1\tSM:Sample1\tCN:SequencingCenter\tDS:Project1\tDT:2015-02-09\tPL:ILLUMINA" -M human_g1k_v37_decoy.fasta read1.fastq.gz read2.fastq.gz

That's likely to work. Also, I added the SM tag, which picard requires. I ran into an issue like this once when I was sending commands through multiple layers of programs on a cluster. It turned out that the "\t" indicators were getting expanded to actual tabs at some point, so I had to escape things ("\\t").

[jmarshall]

There's a bug in bwa here: in main() in main.c it ought to sanitise argv[i] (say, by changing tabs and newlines to spaces) before adding it to the @PG header. (And similarly argv[0] really.)

As dpryan notes, bwa will parse \t escape sequences so this would provide a workaround as \t will appear in the @PG line rather than actual tab characters. Depending on how many layers of scripting are involved, it may or may not be easy to get \t sequences to bwa without a previous layer turning them into tabs first!

[dpryan]

Indeed, BWA will also allow incorrect header tags like "asd:f".

[yinshe]

Thanks, dpryan and jmarshall

I have two layers of scripting. I will try to see whether it is possible to print "\t". It would be best if SAM format has some way to escape tabs in the command line.

I tried to set "VALIDATION_STRINGENCY" as "LENIENT" in picard, it goes through with the warning. I hope the downstream tools can also accept this.