References

The pixy method of dealing with missing data for π calculation was introduced in this paper. You can find discussion on the applicability of the pixy metric here and here.

Our choice of Hudson's Fst estimator (and the way of summarizing it over sites) is influenced by this paper whose authors argue that it is least burdened with assumptions and least biased with variable sample sizes. Usage of Fst estimation for multiple alleles is advocated by Weir and Cockerham here.

In earlier versions, we suggested to parallelize piawka using GNU parallel. It is nice but carefully typing those long commands gets annoying.