New DNA coalescent models and old population genetics software

The analyses of often large amounts of field and laboratory data depend on computer programs to generate descriptive statistics and to test hypotheses. The algorithms in these programs are often complex and can be understood only with advanced training in mathematics and programming, topics that are beyond the capabilities of most fisheries biologists and empirical population geneticists. The backward looking Kingman coalescent model, based on the classic forward-looking Wright-Fisher model of genetic change, is used in many genetics software programs to generate null distributions against which to test hypotheses. An article in this issue by Niwa et al. shows that the assumption of bifurcations at nodes in the Kingman coalescent model is inappropriate for highly fecund Japanese sardines, which have type III life histories. Species with this life history pattern are better modelled with multiple mergers at the nodes of a coalescent gene genealogy. However, only a few software programs allow analysis with multiple-merger coalescent models. This parameter misspecification produces demographic reconstructions that reach too far into the past and greatly overestimates genetically effective population sizes (the number of individuals actually contributing to the next generation). The results of Niwa et al. underline the need to understand the assumptions and model parameters in the software programs used to analyse DNA sequences.