HaplotypeCaller DP reports low values

Dear GATK Team,

I've recently been exploring HaplotypeCaller and noticed that, for my data, it is reporting ~10x lower DP and AD values in comparison to reads visible in the igv browser and reported by the UnifiedGenotyper.

I'm analyzing a human gene panel of amplicon data produced on a MiSeq, 150bp paired end. The coverage is ~5,000x.

My pipeline is:

Novoalign -> GATK (recalibrate quality) -> GATK (re-align) -> HaplotypeCaller/UnifiedGenotyper.

Here are the minimum commands that reproduce the discrepancy:

java -jar /GenomeAnalysisTK-2.7-4-g6f46d11/GenomeAnalysisTK.jar \
-T HaplotypeCaller \
--dbsnp /gatk_bundle/dbsnp_137.hg19.vcf \
-R /gatk_bundle/ucsc.hg19.fasta \
-I sample1.rg.bam \
-o sample1.HC.vcf \
-L ROI.bed \
-dt NONE \
-nct 8

Example variant from sample1.HC.vcf:

chr17 41245466 . G A 18004.77 . AC=2;AF=1.00;AN=2;BaseQRankSum=1.411;ClippingRankSum=-1.211;DP=462;FS=2.564;MLEAC=2;MLEAF=1.00;MQ=70.00;MQ0=0;MQRankSum=0.250;QD=31.14;ReadPosRankSum=1.159 GT:AD:DP:GQ:PL 1/1:3,458:461:99:18033,1286,0

... In comparison to using UnifiedGenotyper with exactly the same alignment file:

java -jar /GenomeAnalysisTK-2.7-4-g6f46d11/GenomeAnalysisTK.jar \
-T UnifiedGenotyper \
--dbsnp /gatk_bundle/dbsnp_137.hg19.vcf \
-R /gatk_bundle/ucsc.hg19.fasta \
-I sample1.rg.bam \
-o sample1.UG.vcf \
-L ROI.bed \
-nct 4 \
-dt NONE \
-glm BOTH

Example variant from sample1.UG.vcf:

chr17 41245466 . G A 140732.77 . AC=2;AF=1.00;AN=2;BaseQRankSum=5.488;DP=6382;Dels=0.00;FS=0.000;HaplotypeScore=568.8569;MLEAC=2;MLEAF=1.00;MQ=70.00;MQ0=0;MQRankSum=0.096;QD=22.05;ReadPosRankSum=0.104 GT:AD:DP:GQ:PL 1/1:56,6300:6378:99:140761,8716,0

I looked at the mapping quality and number of the alignments at the example region (200nt window) listed above and they look good:

awk '{if ($3=="chr17" && $4 > (41245466-100) && $4 < (41245466+100))  print}' sample1.rg.sam | awk '{count[$5]++} END {for(i in count) print count[i], i}' | sort -nr
8764 70
77 0

With other data generated in our lab, that has ~200x coverage and the same assay principle [just more amplicons], the DP reported by HaplotypeCaller corresponds perfectly to UnifiedGenotyper and igv.

Is there an explanation as to why I should see a difference between HaplotypeCaller and UnifiedGenotyper, using these kinds of data?

Many thanks in advance,

Sam