CLUMPP: a cluster matching and permutation program for dealing with label switching and multimodality in analysis of population structure

This paper addresses a practical problem in population-genetic clustering, where individual multilocus genotypes are partitioned into clusters by unsupervised, stochastic algorithms such as BAPS, STRUCTURE and TESS. Replicate cluster analyses of the same data, even from the same initial conditions, may produce several distinct solutions for the estimated cluster membership coefficients. The authors identify two main sources of these differences: label switching, caused by the arbitrary way clusters are labeled in an unsupervised analysis, and genuine multimodality, in which the replicates truly find distinct solutions.

To make clustering results easier to interpret, the authors describe three algorithms for aligning multiple replicate analyses of the same dataset and implement them in the program CLUMPP (CLUster Matching and Permutation Program). They illustrate its use by aligning cluster membership coefficients from 100 replicate cluster analyses of 600 chickens from 20 different breeds. CLUMPP is made freely available, giving researchers a standard way to reconcile replicate clustering solutions affected by label switching and multimodality.