What's new in Version 2?

Perhaps the most obvious addition to structure in Version 2 is a fancy Java front end. The front end includes several new features, producing various plots of data summaries, as well as enabling batch runs, and organizing the data from multiple runs on the same project. C executables will still be available for those who prefer using the command line. See section 8. We have added command line options to simplify simulation studies and batch runs (section 7.6). On the analysis side, the major advance is that we have now implemented a model that allows for ``admixture linkage disequilibrium''. That is, we model the correlations that arise between linked markers as the result of admixture or hybridization. In order to use this, the user must enter map distances (or at least some kind of relative distance) between the markers, and the program then accounts for the possibility of correlations due to linkage. We still do not attempt to model the LD that occurs within populations between very nearby markers ($<1$ cM in humans). See section 4.2. We have also modified and improved the model of correlated allele frequencies, and we find that this model is often more effective than the independent frequencies model at detecting subtle population structure. However, this may come at some cost: we believe that it may be less robust to minor model departures like some inbreeding, or undetected null alleles. This could create a bias towards overestimating the number of clusters, $K$. We have made the new model the default model, but encourage users to experiment with different models. Other new extensions include the possibility of using different $\alpha$s for each population, and estimating $\lambda$s. The computational techniques underlying the new analytical methods are described by Falush et al. (2003). We have also modified the program so that it can be applied to non-diploid organisms. As Daniel Falush points out, this should keep the dandelion population geneticists happy.