I work with non-human genomes and commonly need the confidence of the reference sites, so I was happy to see the inclusion of the RGQ score in the format field of GenotypeGVCFs. However, I am a little confused as to what this score means (how it is calculated). Out of curiosity I plotted the distribution of RGQ and GQ scores over ~1Mbp. A few things jumped out that I was hoping you could explain:
(1) There are two peaks of GQ and RGQ scores, one at 99 - which is obviously just the highest confidence score and another at exactly GQ/RGQ=45. You can see this in the GQ/RGQ distribution below. I've excluded the sites where RGQ/GQ = 0 or 99 (RGQ = blue, GQ=red) is there some reason why so many GT calls == 45?
(2) There are very few GQ = 0 calls and ~96% are GQ=99 - but in the RGQ ~42% == 0 and 54%=99. Is there any explanation why so many RGQ scores == 0? I fear that filtering on RGQ will bias the data against reference calls and include a disproportionate number of variant calls.