Characterization of a Bayesian genetic clustering algorithm based on a Dirichlet process prior and comparison among Bayesian clustering methods

Abstract

BackgroundA Bayesian approach based on a Dirichlet process (DP) prior is useful for inferring genetic population structures because it can infer the number of populations and the assignment of individuals simultaneously. However, the properties of the DP prior method are not well understood, and therefore, the use of this method is relatively uncommon. We characterized the DP prior method to increase its practical use.ResultsFirst, we evaluated the usefulness of the sequentially-allocated merge-split (SAMS) sampler, which is a technique for improving the mixing of Markov chain Monte Carlo algorithms. Although this sampler has been implemented in a preceding program, HWLER, its effectiveness has not been investigated. We showed that this sampler was effective for population structure analysis. Implementation of this sampler was useful with regard to the accuracy of inference and computational time. Second, we examined the effect of a hyperparameter for the prior distribution of allele frequencies and showed that the specification of this parameter was important and could be resolved by considering the parameter as a variable. Third, we compared the DP prior method with other Bayesian clustering methods and showed that the DP prior method was suitable for data sets with unbalanced sample sizes among populations. In contrast, although current popular algorithms for population structure analysis, such as those implemented in STRUCTURE, were suitable for data sets with uniform sample sizes, inferences with these algorithms for unbalanced sample sizes tended to be less accurate than those with the DP prior method.ConclusionsThe clustering method based on the DP prior was found to be useful because it can infer the number of populations and simultaneously assign individuals into populations, and it is suitable for data sets with unbalanced sample sizes among populations. Here we presented a novel program, DPART, that implements the SAMS sampler and can consider the hyperparameter for the prior distribution of allele frequencies to be a variable.

Inter-laboratory evaluation of the EUROFORGEN Global ancestry-informative SNP panel by massively parallel sequencing using the Ion PGM™

“…We opted to use single reference populations from 1000 Genomes without high levels of admixture and with low intra-population variability [26], consisting of AFR (ESN), EUR (GBR) and EAS (JPT) groups plus sets of two CEPH OCE populations and five CEPH AMR populations (Table 1, population no. 1-6).…”

“…4A, orange cells) from population analyses described in section 3.6. This strategy also compensated for the large contrasts in sample size of 1000 Genomes data for the first three groups alongside the much smaller sample sizes of Oceanian and Native American populations, which can interfere with STRUCTURE analyses [26].…”

Identifying Loci Under Selection Against Gene Flow in Isolation-with-Migration Models

“…It results from the fact that if the MCMC chain is mixing properly, all the equivalent indexing assignment vectors will be sampled equally. To summarize the posterior sample in a way that accounts for the uncertainty of assignment of each locus, as well as for label switching, we considered three methods: (1) pairwise coassignment probabilities (e.g., Dawson and Belkhir 2001;Onogi et al 2011), (2) mean assignment based on partition distances (Huelsenbeck and Andolfatto 2007), and (3) marginal probabilities after relabeling [e.g., pivotal reordering (Lee et al 2009)]. With the first approach it is possible to identify, using Bayes factors, those pairs of loci for which there is strong evidence that both loci belong to the same group.…”

“…Mean assignment: Although it would not be meaningful to take an arithmetic average of sampled assignments, we can use the idea of "mean assignment", a, which is the vector that minimizes the squared partition distance to all of the sampled vectors (Huelsenbeck and Andolfatto 2007;Choi and Hey 2011;Onogi et al 2011). The partition distance is defined as the minimum number of elements that must be removed to make two assignment vectors identical, over all possible labeling of one of the vectors (Gusfield 2002).…”

“…Because the minimum is taken over possible labelings, the mean assignment is not sensitive to label switching. The mean assignment computation was implemented by adapting the mean partition functions from Onogi et al (2011) and Choi and Hey (2011).…”