bam file and vcf wildly disagree

[Mark]

Hi All

If anyone can explain this I would appreciate it.

I ran blasr on some pacbio ccs data in fasta format. I then attempted to use freebayes to call variants on the resulting bam file.

freebayes  -F 0.5 --pooled-continuous -f ref.fa  -C 3 pacbio_ccs_reads.fa-vs-ref.fa.sort.bam >pacbio_ccs_reads.fa-vs-ref.fa.sort.bam.vcf

A number of variants were called. The trouble is that I see little relation between what is reported in the vcf file and the bam file as viewed in Tablet (see attached snippet of the alignment at this example position). For example:

#CHROM  POS     ID      REF     ALT     QUAL    FILTER  INFO    FORMAT  NO_CHIP_ID
ref      3659    .       TAT     TT      15.2798 .       AB=0;ABP=0;AC=2;AF=1;AN=2;AO=4;CIGAR=1M1D1M;DP=7;DPB=7;DPRA=0;EPP=3.0103;EPPR=0;GTI=0;LEN=1;MEANALT=4;MQM=254;MQMR=0;NS=1;NUMALT=1;ODDS=3.48821;PAIRED=0;PAIREDR=0;PAO=0;PQA=0;PQR=-34;PRO=1;QA=0;QR=0;RO=0;RPL=1;RPP=5.18177;RPPR=0;RPR=3;RUN=1;SAF=3;SAP=5.18177;SAR=1;SRF=0;SRP=0;SRR=0;TYPE=del GT:[B]DP[/B]:DPR:[B]RO[/B]:QR:[B]AO[/B]:QA:GL        1/1:[B]7[/B]:7,4:[B]0[/B]:0:[B]4[/B]:0:-3.145,-4.65015,0

In tablet, four 1bp deletions are indeed seen at this position (as reported in the vcf), but the coverage depth is 115 (not 7), all but 6 of which are the reference. In the above vcf, reference occurence is 0. In fact, reference occurence for all called variants is 0, which is clearly not true in the bam file, and coverage is much higher in the bam than ever report in the vcf.

Any ideas?

Thanks

Addendum:
Now if I do the same thing with the run's subreads (fastq format) I get a freebayes result that is much more in keeping with what I see in the bam alignment data (though not exact, at least it reflects the low variant occurence relative to the reference occurence). WTF.