Strategy for processing ancient DNA

Question

[Ghani]

Hi everyone,

My favorite caller with regards to true and false positive variant results is HaplotypeCaller combined with BWA mem for mapping. This is based on personal experience using Freebayes and bcftools, but I have not compared haplotypecaller with callers such as STSchiff which is specifically designed for processing aDNA.

Has anyone compared processing low coverage ancient DNA (avg coverage <1) with low endogenous content using HaplotypeCaller and stschiff or similar. If so, what settings did you use with HaplotypeCaller to minimize reference bias at very low coverage loci, and how was the comparison between callers.

[Ghani]

With regards to optimizing settings on HaplotypeCaller:

Besides setting --minPruning and --minDanglingBranchLength to 1, so as to not throw out any branch of the graph that has less than 2 reads during realignment, since default minPruning is 2, and since length of a dangling branch to attempt recovery is 4 by default, is there anything else that can be done.

What about lowering the emit confidence threshold, and call confidence thresholds also. Any suggested values lower than default? Or is this unnecessary.

Also was wondering if the false positives will be slightly higher or significantly higher using the settings in my 1st paragraph. I did not see a comparison study. I can live with an increase of FP of about 10%.

Thanks

Answer 1

[the_smiao]

Hi @Ghani,

Don't use HaplotypeCaller with aDNA, ti is optimized and designed to process 30X + datasets. You can experiment with HaplotypeCaller, however, with less than 1X on average, stick with calling pseudo-haploids using random draw. You simply can't call reliably diploid calls, and lowering the thresholds these tools operate under is only going to lower considerably your sensitivity.

To call pseudo-haploids, use samtools 0.1.19, or check sequenceTools by Stephan Schieffels.