Quality in Pileup depends on Reference

[japuch]

Dear all,

i have a strange (for me) problem with generating variant calls by 'samtools mpileup' from a BAM file (attached in archive) generated with novoalign. Namely the base qualities are wrong (different as reported in BAM/SAM) when the reference is provided. Running the following command on the attached BAM:

'samtools mpileup -r chr14:68272021-68272021 test.bam'

yields:
chr14 68272021 N 23 acaCcCacacAAAACCCCACAC^gC ___QFF___]][_\___X\___W

but running:

'samtools mpileup -r chr14:68272021-68272021 -f hg19.fa test.bam'

yields:

chr14 68272021 C 23 a,a.,.a,a,AAAA....A.A.^g. !^!QFF!^!]!%!$^N^X$^!^A

The base quality values in the latter are different than those stored in BAM/SAM.

Any help will be appreciated.

Best Regards,

Jacek

[me_myself_andI]

Hi,

this is most likely due to samtools BAQ computation. Briefly, it will downgrade base-qualities if a misalignment/indel is likely and that will change depending on the reference. See 2021320865. If you don't want this behaviour, you can switch off BAQ computation in samtools mpileup with the option -B, which is however not recommended.

Andreas

[japuch]

Hi,

you were right, thanks. Setting the B parameter recovers the correct behavior. The trick is, however, that none of the reads included in the bam file I attached to my first message has an insertion or deletion (no D or I in CIGAR). That's something I don't understand.


Jacek

[swbarnes2]

Your pileups don't show any indels either, so I don't see what the problem there is.

^g does not mean an insertion of a G.

Also at the read base column, a symbol `^' marks the start of a read segment which is a contiguous subsequence on the read separated by `N/S/H' CIGAR operations. The ASCII of the character following `^' minus 33 gives the mapping quality.

[japuch]

The thing is that I don't understand why the qualities of the bases indicating a SNP in the position mentioned are reduced by BAQ (using BAQ or not leads to different calls at the position). The BAQ should "reduce false SNP calls caused by misalignments around insertions and deletions (indels).", as the paper about BAQ says. As there is no indel around the position mentioned (and thus no reason for misalignment), the BAQ should not change qualities in this case. Or do I misunderstand something here?

Thanks for the help!

Jacek

[swbarnes2]

Just don't use BAQ. No matter what it's supposed to do, it often does just what you observed. Its supposed to reduce the number of false positives, but sometimes, it ruins real positives.