A Bayesian MCMC Implementation of AMOVA

As technologies have improved, the availability of genetic data has increased rapidly. Yet many population genetics studies still use the statistics developed fifty years ago. For example, simple F statistics are inappropriate for loci with many alleles. These statistics do not have a way to reflect confidence in their estimates, nor sources of error in data gathering. We have developed a Bayesian Markov Chain Monte Carlo implementation of AMOVA (Analysis of Molecular Variance), for many loci and alleles. The AMOVA gives parameters analogous to F statistics, called -statistics, which describe population differentiation. The method also gives confidence intervals for the statistics, which incorporate various sources of error in the data. The Bayesian AMOVA technique gives similar results to other programs for individual loci, but can utilize genetic data in which populations rather than individuals are the unit of sampling, and results in a better global estimate of the -statistics across loci than the weighted mean that has been used in the past.