Zygosity difference

[oyvindbusk]

I am currrently looking into some exome data that we have run on our hiscanSQ. We have sequenced a sample in two different runs, to compare the number of identified variants in both runs.

I also looked into the zygosity of the samples, and found that 16 % (!!) of the variants differed in their zygosity (e.g. hetozygous in one run, heterozygous in the next). This sounds high, so I was wondering if anybody else has looked into this in their data, and want to share their experience?.

Details of analysis:
I used a pipeline with bwa + GATK for alignment and genotyping. Initially, I filtered my vcf-files so that only variants covered by at least 30 X was included, and also removed some low-quality variants not passing filters. Then I intersected the two files with BEDtools and added the -wo option, to get both outputs in the same file. I used a simple awk to output the lines in which the GT-field was different.

[Bukowski]

I used a pipeline with bwa + GATK for alignment and genotyping. Initially, I filtered my vcf-files so that only variants covered by at least 30 X was included, and also removed some low-quality variants not passing filters. Then I intersected the two files with BEDtools and added the -wo option, to get both outputs in the same file. I used a simple awk to output the lines in which the GT-field was different.

As I first read this I assumed you hadn't filtered for low coverage variants, but you have. What filters did you use to filter out 'low quality' variants?

Also would vcf-tools perhaps not been a little more straightforward for vcf/vcf comparisons?

[oyvindbusk]

I used these filters in the GATK:
QD < 2.0
MQ < 40.0
FS > 60.0
HaplotypeScore > 13.0
MQRankSum < -12.5
ReadPosRankSum < -8.0

You are probably right that vcftools or VariantEval in GATK would be better to use.

[Bukowski]

Hmm very similar to my exome filtering procedure. I have just checked some samples that have gone through in duplicate on different runs, and I don't see anything like your numbers in my runs, in fact of about 30k variants, only a couple of hundred altered zygosity between runs, and I bet they're all low coverage ones. Whether you pick up the same variants in the runs is another matter, I get >70% concordance for the het calls (i.e. present in both samples).