Bayesian StairwayPlot for Inferring Single Population Demographic Histories From Site Frequency Spectra

Area of Science:

• Population genetics
• Computational biology
• Evolutionary modeling

Background:

• Estimating complex demographic histories is crucial for understanding population evolution.
• The StairwayPlot method offers a flexible approach using site frequency spectrum data.
• Bayesian inference provides a powerful framework for demographic modeling.

Purpose of the Study:

• To implement the StairwayPlot method within the Bayesian software RevBayes.
• To compare the performance of the Bayesian implementation against existing Maximum Likelihood methods.
• To evaluate different prior distributions for population size changes and assess model selection strategies.

Main Methods:

• Utilized expected coalescent times and a multinomial likelihood function.
• Incorporated Bayesian Skyline Plot approaches with various prior distributions (i.i.d., GMRF, HMRF).
• Implemented leave-one-out cross-validation for model selection.

Main Results:

• The Bayesian StairwayPlot in RevBayes demonstrates comparable parameter accuracy to StairwayPlot2.
• Gaussian Markov Random Fields (GMRF) priors best model smoothly varying histories.
• Horseshoe Markov Random Fields (HMRF) priors are optimal for abruptly changing demographic histories.
• As few as 10 diploid individuals and 500k SNPs are sufficient for complex demographic inference.

Conclusions:

• The Bayesian implementation of StairwayPlot in RevBayes is a robust tool for demographic inference.
• Prior choice significantly impacts the accuracy of demographic history reconstruction.
• Empirical studies require careful consideration of sequence length, mutation rate, and potential biases.