The BEST website provides example files for analyses when single alleles are sampled for each species, as well as the sampling of multiple alleles from each species. Both analyses assume that species are reciprocally monophyletic. Given that BEST is a modification of MrBayes, the data formats are very similar except that BEST includes priors for theta and mu. Also, the tree topologies, branch lengths, and mu are unlinked across the sampled loci. If a haploid locus (mitochondrial or chloroplast DNA) is sampled, BEST allows the user to define the ploidy of the locus (default setting is diploid).
If a multiple locus dataset is run in BEST, a sham file is needed to summarize the trees after the burnin (the familiar "sumt" command from MrBayes). Liang Liu has posted an example of this type of file on the BEST website.
The trees are summarized using a burnin value that discards all trees and parameter values sampled prior to convergence. As in MrBayes, summarizing the trees produces a consensus tree file, where the consensus percentages for clades are interpreted as the Bayesian posterior probability. Progress of the BEST run and assessment of convergence can be monitored using the computer program Tracer.
My laboratory group has been experimenting with BEST for the past few months, and we are generating some interesting and exciting results. The prior on theta appears to be the one issue/nuisance that we have run across in our explorations using BEST. A fairly wide prior is given in the example files. We are beginning to run BEST with more narrow, and realistic, priors for theta. So far the results are promising.
Overall, I have found BEST straightforward to implement with my multilocus phylogenetic data. Familiarity with MrBayes will certainly help new users of BEST. Also, Liang Liu has been very helpful and encouraging to users, and has implemented suggestions into the example files on the BEST website. My entire lab group is excited to be exploring the frontier of phylogenetics, with the hope of that we are making the most reasonable inferences regarding species relationships that is afforded by our hard earned data.