On making precision-recall curves: quality info? lenient parameters?

[jmonlong]

I'm benchmarking SVTyper and making PR curves. I have two questions to make sure I'm comparing SVTyper as I should.

I'm using the QUAL field to make the curve. I tried using the GQ field but I'm getting "worse" curves (see below). Any recommendation about which quality information I should use?

Also I was wondering if there was a way to "extend" the curve. Now it stops at a low recall/high precision. Is there a way to have get lower quality genotypes that would "continue" the curve toward a high recall/low precision? Maybe a parameter to run SVTyper with? I didn't see any obvious ones in svtyper -h.

Thanks for your help