A hierarchical Bayesian approach for estimating the origin of a mixed population

Feng Guo, Dipak K. Dey, Kent E. HolsingerInstitute 2008 · DOI · arXiv · OpenAlex

Abstract: We propose a hierarchical Bayesian model to estimate the proportional contribution of source populations to a newly founded colony. By integrating genetic data with environmental and demographic factors, we estimate an assortative mating coefficient, mixture proportions, and regression relationships in a single framework. The model introduces a novel Dirichlet-Dirichlet prior structure to improve the precision of covariate effect estimates. Application to grey seal data from the Orkney Islands demonstrates the model's ability to better isolate environmental influences compared to previous methods.

Introduction

Determining the origins of individuals in a newly established colony is a fundamental challenge in conservation biology. While genetic markers provide essential clues, they are often insufficient on their own due to low population divergence or small sample sizes. This paper, originally published in IMS Collections (2008), introduces a hierarchical Bayesian framework that leverages both genetic data and environmental covariates—such as geographic distance or habitat quality—to provide more robust and interpretable estimates of source contributions.

The Hierarchical Model

The proposed model operates through several interconnected layers to handle the complexity of biological migration. The first layer defines the likelihood of observed genotypes in the new colony using a mixture multinomial distribution. A critical addition here is the assortative mating coefficient ($\omega$), which accounts for the biological reality that new migrants may prefer to mate with others from their original source population rather than mixing randomly.

The second layer introduces the model's primary innovation: a Dirichlet-Dirichlet prior structure. In this setup, the mixture proportions ($m$) are drawn from a Dirichlet distribution, whose parameters are themselves governed by a second Dirichlet distribution ($\phi$) and a dispersion parameter ($\rho$).

Directed Acyclic Graph (DAG) of the proposed Dirichlet-Dirichlet hierarchical model, showing the relationships between the genetic data (G), environmental factors, and population parameters.

Innovation: Separating Signal from Noise

The shift to a Dirichlet-Dirichlet structure addresses a major flaw in older "Dirichlet-Lognormal" models. In previous frameworks, the mean effect of an environmental covariate was often confounded with the variance (uncertainty) in the data. This led to "blurry" estimates where it was difficult to tell if a factor like distance was truly influential or if the data was simply noisy.

By nesting two Dirichlet distributions, the authors isolate the "average effect" of a covariate from the "noise." The parameter $\rho$ acts as a built-in indicator of goodness-of-fit: if $1-\rho$ is high, it signifies that the chosen environmental factors (like distance) are doing an excellent job of explaining the observed population mix.

Application: Grey Seals in the Orkney Islands

The model was validated using real-world data from grey seal colonies in the Orkney Islands, Scotland. Researchers sought to determine the source of a newly established colony on Stronsay island using eight genetic loci from seven potential source colonies. They integrated two specific environmental covariates: the geographic distance to the new colony and a productivity index representing habitat quality.

Table of environmental covariates (Distance and Productivity) for the seven source populations used in the grey seal study.

Key Results and Insights

  • Precision Gains: Both simulations and the seal study demonstrated that the Dirichlet-Dirichlet model produced significantly tighter credible intervals for regression coefficients compared to previous methods. This makes the results much more useful for conservation decision-making.
  • Environmental Drivers: The analysis of the Orkney seals revealed that distance had a moderate negative effect—seals are more likely to colonize nearby islands. Interestingly, the productivity of the source island had a negligible effect on the likelihood of migration.
  • Data Quality: Simulations confirmed that while natural population divergence is fixed, researchers can significantly improve model performance by increasing the number of genetic markers (loci) analyzed.

Conclusion

This hierarchical Bayesian approach provides a sophisticated tool for ecological inference. While the small number of source populations in the seal study limited statistical power, the framework is designed to scale. It offers a mathematically rigorous way to combine "common sense" environmental data with messy genetic evidence, providing a clearer picture of how new populations are formed and maintained.

Summary updated

Abstract

<!-- *** Custom HTML *** --> We propose a hierarchical Bayesian model to estimate the proportional contribution of source populations to a newly founded colony. Samples are derived from the first generation offspring in the colony, but mating may occur preferentially among migrants from the same source population. Genotypes of the newly founded colony and source populations are used to estimate the mixture proportions, and the mixture proportions are related to environmental and demographic factors that might affect the colonizing process. We estimate an assortative mating coefficient, mixture proportions, and regression relationships between environmental factors and the mixture proportions in a single hierarchical model. The first-stage likelihood for genotypes in the newly founded colony is a mixture multinomial distribution reflecting the colonizing process. The environmental and demographic data are incorporated into the model through a hierarchical prior structure. A simulation study is conducted to investigate the performance of the model by using different levels of population divergence and number of genetic markers included in the analysis. We use Markov chain Monte Carlo (MCMC) simulation to conduct inference for the posterior distributions of model parameters. We apply the model to a data set derived from grey seals in the Orkney Islands, Scotland. We compare our model with a similar model previously used to analyze these data. The results from both the simulation and application to real data indicate that our model provides better estimates for the covariate effects.
BibTeX
@inproceedings{guo2008hierarchical,
  author = {Feng Guo and Dipak K. Dey and Kent E. Holsinger},
  title = {{A hierarchical Bayesian approach for estimating the origin of a mixed population}},
  booktitle = {Institute},
  year = {2008},
  doi = {10.1214/074921708000000174},
}
Figure