Hello all, While benchmarking GIAB data with sarek (DeepVariant and HaplotypeCaller), we observed several false positives occurring at homopolymer boundaries. GIAB confirmed this is a common source of errors and suggested we ask here about best practices. Do you have recommended filtering strategies within Sarek (or downstream QC) to handle homopolymer-related FPs for DV and HC? Example variants: • chr1:115725557 GAAA>G • chr10:87960876 C>CT • chr3:10152060 CA>C • chr9:132897613 G>GAA Thanks in advance for any guidance.

Deepvariant is something that could be trained to ignore such artifacts. I have no experience with it, however, and don't know how to get that into sarek. For finding/filtering all homopolymer variants you can intersect the variants with some annotation of homopolymer regions. This likely will remove true positives as well, but if you have a way of distinguishing those, you could use that criterion on the filtered variants. AI thinks that SV-Pop can do that, but I have never used that. Alternatively, if you can't find any homopolymer annotation, you could write a script that scans a genome FASTA and makes a BED file entry whenever it find 4+ bases of the same kind. When you intersect that with the variant VCF (e.g. using

bedtools

), make sure that you are not missing insertion variants where the base at

POS

in the VCF is not part of the homopolymer in the reference. (Not sure about the best way of doing this -- maybe some padding of the intervals in the BED?)