Mutect2 occasionally writes nonsense / invalid values for MPOS info tag

[daviesrob]

Bug Report

Affected tool(s) or class(es)

Mutect2

Affected version(s)

• Latest public release version [4.1.4.1]
• Latest master branch as of [date of test?]

Description

Mutect2 occasionally writes lines including INFO tag MPOS=-2147483648. This doesn't look sensible for "median distance from end of read", and the specific value is disallowed in section 1.3 of the VCF specification.

I've had a quick look at the code, and think the dubious value may be generated in ReadPosition::getValueForRead when the result from ReadPosRankSumTest.getReadPosition is cast to an int. Looking at that function, it can return INVALID_ELEMENT_FROM_READ which is defined as Double.NEGATIVE_INFINITY. According to the java documentation, casting NEGATIVE_INFINITY to int will result in a value of INT_MIN. (Disclaimer: I haven't tested this, so it may be completely wrong...)

Steps to reproduce

See attached .zip file which includes a smallish bam file that shows the problem. I ran mutect2 on it in the Docker container for the latest GATK release:

unzip mpos_issue.zip
cd mpos_issue
../gatk Mutect2 --input input/small.bam --reference input/small.fa --output small.vcf
grep MPOS=- small.vcf

Expected behavior

MPOS should have a sensible value.

Actual behavior

ref|NC_001224|	12835	.	A	AATAT	.	.	DP=2810;ECNT=5;MBQ=20,34;MFRL=185,202;MMQ=60,29;MPOS=-2147483648;POPAF=7.30;TLOD=3.04	GT:AD:AF:DP:F1R2:F2R1:PGT:PID:PS:SB	0|1:2804,3:1.431e-03:2807:1316,3:1412,0:0|1:12828_AATAC_A:12828:1200,1604,2,1

---

mpos_issue.zip

[adrlar]

I am having an issue with the same bug, where these MPOS=-2147483648 fields are causing an error in downstream analysis. If anyone has any update or a workaround I'd appreciate it.

[DrJCampbell]

I'm also having a problem with this bug breaking downstream analysis. Using bcftools to index the vcf containing MPOS=-2147483648 converts it to MPOS=..