I've mapped ~900 samples on my reference (non-model species) and I'd like to have some stats about the quality, like Fst, heterozygosity, look how they pan into a PCA.

I have sampled 100 regions of 25kbp and SNP called them, then I converted my BCF into VCF.

Now I am wondering if I should treat it like a "final" VCF file, hence soft-filter for bad quality calls, sites near indels and bad calls. I surely need to remove NA for the PCA, and this tells me I'll have to inpute as well.

It's a cursory view, so I'm a bit concerned of overkilling. What is the good practice?