Overlapping INDELs in HaplotypeCaller output (GATK v3.4-46)

I am confused by the following output which is generated from HaplotypeCaller. You can see win the snip below that there are 2 deletions 4 bases apart which significantly overlap each other.

CHROM POS ID REF ALT QUAL FILTER INFO FORMAT

3 53087898 . AAAAAAAAAAAATATAT A 106.04 PASS AC=2;AF=0.500;AN=4;BaseQRankSum=-3.900e-02;ClippingRankSum=-1.752e+00;DP=45;FS=0.000;MLEAC=1;MLEAF=0.250;MQ=60.00;MQRankSum=-1.950e-01;QD=5.05;ReadPosRankSum=1.52;SOR=0.368 GT:AD:DP:GQ:PL 0/1:10,0:12:2:2,0,279 0/1:15,6:21:99:127,0,786
3 53087902 . AAAAAAAATATATATAT A 38.47 PASS AC=1;AF=0.250;AN=4;BaseQRankSum=1.23;ClippingRankSum=1.23;DP=43;FS=2.682;MLEAC=1;MLEAF=0.250;MQ=60.00;MQRankSum=-3.580e-01;QD=7.69;R.eadPosRankSum=1.62;SOR=1.863 GT:AD:DP:GQ:PL 0/1:2,3:5:68:68,0,149 0/0:16,3:19:13:0,13,633

My questions:
1. How should I physically interpret this output?
2. Is the higher quality variant the preferred haplotype in this case?

I note a similar issue was reported in this post with inserts, where this behaviour was confirmed as 'expected' but without explanation of the physical meaning of this overlap: http://gatkforums.broadinstitute.org/gatk/discussion/6307/confused-by-overlapping-indels

The call which produced the output was:

gatk -T HaplotypeCaller -R ${ref} -I ${bam} -o ${out}/haplotype_call.g.vcf --dbsnp ${calling_dbsnp} --emitRefConfidence GVCF -stand_call_conf 30 -stand_emit_conf 15 -ploidy 2

followed by

gatk -T GenotypeGVCFs -R ${ref} --variant ${out}/haplotype_call.g.vcf -o ${out}/variants_raw.vcf --dbsnp ${calling_dbsnp}

Answers

  • SheilaSheila Broad InstituteMember, Broadie, Moderator admin

    @peterpriestley
    Hi,

    Can you please post the original bam file and bamout file for the two samples at the site?

    Also, can you please try running the latest version (3.5) of HaplotypeCaller and GenotypeGVCFs on that region and confirm the output looks the same? You can just run on 3:53087598-53088198.

    Thanks,
    Sheila

Sign In or Register to comment.