The hitchhiker's guide to single-locus species delimitation

Molecular approaches to species delimitation are increasingly used to ascertain the number of species in a sample prior to taxonomic, ecological or physiological studies. Although multilocus approaches are gaining fast in popularity, single-gene methods still predominate in the literature. However, available simulation benchmarks of these methods focus exclusively on species-poor samples and/or tree-based approaches: as a result, travellers in the land of single-locus species delimitation lack a comprehensive "hitchhiker's guide" highlighting the sweet spots and dangers on their road. To fill this gap, we compared the performances of distance-based (ABGD, "automatic barcode gap discovery"), allele sharing-based (haplowebs) and tree-based approaches (GMYC, "generalized mixed Yule-coalescent" and PTP, "Poisson tree processes") to detect interspecific boundaries in samples of 6, 60 and 120 simulated species with various speciation rates, effective population sizes, mutation rates and sampling patterns. We found that all approaches performed poorly when population sizes and speciation rates were large, with haplowebs yielding best results followed by ABGD then tree-based approaches. The latter's error type was mostly oversplitting, whereas ABGD chiefly overlumped and haplowebs leaned either way depending on simulation parameters: such widely divergent error patterns suggest that, if all three types of methods agree, then the resulting delimitation is probably correct. Perfect congruence being quite rare, travellers in search of a one-size-fit-all approach to single-locus species delimitation should forget it; however, our hitchhiker's guide raises hope that such species delimitation's Holy Grail may be found in the relatively uncharted nearby land of multilocus species delimitation.