Related Experiment Videos
Comparing likelihood and Bayesian coalescent estimation of population parameters.
Mary K Kuhner1, Lucian P Smith
1Department of Genome Sciences, University of Washington, Seattle, Washington 98195, USA. mkkuhner@gs.washington.edu
Genetics
|March 3, 2006
Summary
We developed a Bayesian genealogy sampler and found it comparable to the likelihood version for estimating population genetics parameters. The Bayesian method showed fewer rejections of true values, suggesting improved accuracy in parameter estimation.
Area of Science:
- Population Genetics
- Computational Biology
- Statistical Inference
Background:
- Markov chain Monte Carlo (MCMC) methods are crucial for inferring demographic history from genetic data.
- Existing likelihood-based samplers, like LAMARC, provide valuable insights into population genetics parameters.
- The integration of Bayesian inference offers a complementary framework for parameter estimation and uncertainty quantification.
Purpose of the Study:
- To develop and evaluate a Bayesian version of the LAMARC genealogy sampler.
- To compare the performance of Bayesian and likelihood-based MCMC approaches for estimating key population genetics parameters.
- To assess the accuracy of parameter estimates, including population size (theta = 4N(e)mu), exponential growth rate, and recombination rate, using simulated DNA data.
Main Methods:
- Development of a Bayesian implementation of the LAMARC genealogy sampler.
- Utilized simulated DNA datasets to rigorously test both Bayesian and likelihood-based MCMC methods.
- Assessed the accuracy of estimated parameter means and the reliability of support or credibility intervals.
Main Results:
- Both Bayesian and likelihood-based methods yielded highly similar results for parameter estimation.
- Both approaches occasionally produced overly narrow intervals, leading to the exclusion of true parameter values.
- The Bayesian approach demonstrated a significantly lower rate of rejecting true generative parameter values compared to the likelihood approach.
Conclusions:
- The Bayesian genealogy sampler performs comparably to its likelihood-based counterpart in estimating population genetics parameters.
- The Bayesian framework offers an advantage by reducing the frequency of rejecting true parameter values, potentially improving inference reliability.
- Further investigation into interval width issues is warranted for both methods to enhance the precision of demographic inference.