We have a 15bp deletion in a sample that was run on an Illumina sequencer twice:
1) on GAII, 76bp paired-end reads, 10-sample pool;
2) HiSeq2000, 100bp paired-end reads, 24-sample pool.
Both times aligned with bwa and analysed with recommended GATK protocol using the latest available version. The deletion was called both times by GATK 1.0.5083 (76bp batch) and GATK 1.4 (100bp batch). However the deletion is not detected by GATK 2.0 or 2.1-8, either by UnifiedGenotyper or HaplotypeCaller. The newer GATK versions do generate fewer calls and substantially cut out false positives, but on the other hand the sensitivity seems to have dropped too much! Here is the variant called by the GATK 1.4, 100bp batch:
7 44190538 . CCCTCCACCCGGCCCA C 8793.94 PASS AC=1;AF=0.021;AN=48;BaseQRankSum=14.575;DP=7878;FS=10.160;HRun=0;HaplotypeScore=3737.9770;InbreedingCoeff=-0.0213;MQ=57.74;MQ0=0;MQRankSum=-10.741;QD=29.71;ReadPosRankSum=-0.056 GT:AD:DP:GQ:PL 0/1:216,80:296:99:8794,0,39333
And here it is called by GATK 1.0.5083, 76bp batch:
7 44190538 . CCCTCCACCCGGCCCA C 7765.65 PASS AC=1;AF=0.050;AN=20;DP=1907;Dels=0.02;HRun=1;HaplotypeScore=40.0818;MQ=59.30;MQ0=0;QD=49.78;SB=-1238.06;sumGLbyD=49.78 GT:DP:GQ:PL 0/1:118:99:7766,0,14618
Where is this loss in sensitivity likely to be occurring? Can we adjust any of the default GATK settings to be able to detect these slightly larger indels? Reads with the 15bp deletion inherently have slightly lower mapping qualities, but I would have thought this is taken into account for indel calling. (I should add that we have a sample with a 27bp homozygous deletion and this always gets called.)
I'd be happy to send the relevant part of the bam file if that would help.