Identifying the number of population clusters with structure: problems and solutions

Abstract

The program structure has been used extensively to understand and visualize population genetic structure. It is one of the most commonly used clustering algorithms, cited over 11,500 times in Web of Science since its introduction in 2000. The method estimates ancestry proportions to assign individuals to clusters, and post hoc analyses of results may indicate the most likely number of clusters, or populations, on the landscape. However, as has been shown in this issue of Molecular Ecology Resources by Puechmaille (), when sampling is uneven across populations or across hierarchical levels of population structure, these post hoc analyses can be inaccurate and identify an incorrect number of population clusters. To solve this problem, Puechmaille () presents strategies for subsampling and new analysis methods that are robust to uneven sampling to improve inferences of the number of population clusters.