This site is now read-only. You can find our new documentation site and support forum for posting questions here.
Be sure to read our welcome blog!
HaplotypeCaller and downsampling
What I have learned so far from other discussions about HaplotypeCaller:
- read counts for positions with very high coverage are downsampled
- this does not affect variant calling
- this does affect DP and AD fields in the output (g)vcf file
- reads are selected randomly
- don't use -nct parameter with HC
- downsampling is hard-coded and can't be influenced by parameters
Nonetheless two problems remain:
The HC doc says "This tool applies the following downsampling settings by default. To coverage: 500" Why is it possible to observe much higher coverage (DP, AD) values in the output vcf file?
I observe SNPs where the recalibrated bam file in IGV has a depth of 1385 for the reference and 1233 for alternate allele but 839 (reference) and 246 (alt) in the HaplotypeCaller vcf file. Maybe this happens by chance, as reads for downsampling are chosen at random or it is related to this bug [gatkforums.broadinstitute.org/discussion/5882/uncorrect-strand-bias-due-to-downsampling-haplotypecaller](http://gatkforums.broadinstitute.org/discussion/5882/uncorrect-strand-bias-due-to-downsampling-haplotypecaller
Both observations lead to the conclusion that DP and AD values from HC output are of little use for samples with high (where does high start? 500?) coverage.