Source-linked AI summary
Inferring population history with DIYABC: a user-friendly approach to Approximate Bayesian Computation
Jean-Marie Cornuet, Filipe Santos, Mark A. Beaumont, Christian P. Robert, Jean-Michel Marin, David J. Balding, Thomas Guillemaud, Arnaud Estoup
TL;DR
Population-genetic inference tools have largely focused on simple scenarios, despite empirical samples often reflecting more complex histories. DIYABC applies customizable Approximate Bayesian Computation to compare scenarios, estimate parameters, and assess estimation confidence, with simulated analyses illustrating these capabilities.
Problem
Existing population-genetic inference methods and software mainly address simple standard evolutionary scenarios, whereas collected samples often do not fit them.
Method
DIYABC uses Approximate Bayesian Computation by simulating genetic data from prior-drawn parameters, comparing summary statistics with observed data, and applying regression-based posterior estimation.
Results
Scenario 1 was significantly better supported than the alternatives, with logistic regression and direct approaches giving congruent results.
Takeaways & Limitations
DIYABC enables nonspecialists to analyze complex population-evolutionary scenarios, compare competing scenarios, estimate parameters, and evaluate estimation confidence.
Takeaways & Limitations
Very small population sizes can produce multiple coalescences in one generation, including multifurcating genealogies that complicate genealogy simulation.
Abstract
from arXiv · showhide
Genetic data obtained on population samples convey information about their evolutionary history. Inference methods can extract this information (at least partially) but they require sophisticated statistical techniques that have been made available to the biologist community (through computer programs) only for simple and standard situations typically involving a small number of samples. We propose here a computer program (DIYABC) for inference based on Approximate Bayesian Computation (ABC), in which scenarios can be customized by the user to fit many complex situations involving any number of populations and samples. Such scenarios involve any combination of population divergences, admixtures and stepwise population size changes. DIYABC can be used to compare competing scenarios, estimate parameters for one or more scenarios, and compute bias and precision measures for a given scenario and known values of parameters (the current version applies to unlinked microsatellite data). This article describes key methods used in the program and provides its main features. The analysis of one simulated and one real data set, both with complex evolutionary scenarios, illustrates the main possibilities of DIYABC
Summary:
The paper identifies its authors and states that DIY ABC is freely available.
- DIY ABC is freely available online.
- The paper was published by Oxford University Press in 2008.
1 INTRODUCTION
Population-genetic inference software has largely targeted simple evolutionary scenarios, whereas real samples often require complex histories. The paper motivates ABC and DIY ABC as an accessible way to analyze such situations and assess scenario and parameter uncertainty.
- Most existing population-genetic inference methods and software address simple standard scenarios involving one to three populations.
- Real samples often do not match these standard scenarios, forcing researchers either to discard samples or develop specialized software.
- Complex scenarios create computational difficulties for likelihood-based approaches, motivating Approximate Bayesian Computation, or ABC.
- ABC compares simulated and observed data through similarity of summary statistics rather than directly computing the likelihood.
- DIY ABC was developed to make ABC accessible for complex scenarios and to estimate scenario support, parameter uncertainty, and bias and precision.
2 KEY METHODS INVOLVED IN DIYABC
DIYABC implements ABC through simulation, distance-based rejection, and regression adjustment, while modular genealogy simulation supports divergence, admixture, repeated sampling, and population-size variation. It also compares scenarios using direct proportions or weighted logistic regression and switches coalescence algorithms to balance speed and bias.
- ABC inference: ABC first builds a reference table of prior-drawn parameters, simulated data sets, and their summary statistics.
- ABC inference: DIYABC retains simulations closest to the observed summary statistics, using selected parameter values to approximate posterior distributions.
- Genealogical simulation: Its modular genealogy simulator supports isolated coalescence, lineage pooling for divergence, lineage splitting for admixture, repeated sampling, and population-size variation.
- Coalescence simulation: The continuous-time algorithm is faster in most cases, but DIYABC switches to generation-by-generation simulation when continuous-time approximation becomes unacceptable.
- Coalescence simulation: The switching criterion was calibrated to optimize computation speed while keeping relative bias in coalescence rates below 5%.
- Comparing scenarios: Scenario probabilities are estimated either from proportions among the closest simulations or with weighted polychotomous logistic regression.
3 DIYABC: A COMPUTER PROGRAM FOR POPULATION BIOLOGISTS
DIYABC applies Approximate Bayesian Computation to complex population-genetic scenarios, supporting scenario comparison, parameter estimation, and bias and precision assessment. Its examples illustrate these capabilities on simulated and real microsatellite data.
- Program scope: DIYABC performs ABC inference on diploid microsatellite genotypes, quantifying support for evolutionary scenarios and estimating posterior parameter densities.Missing data are allowed, and the program provides a graphical interface for Windows.
- Program scope: Scenarios may include any number of populations and samples, timed divergences, admixtures, stepwise population-size changes, unsampled populations, and serial sampling.The main restriction is that migration between populations is not supported.
- Program scope: The program supports microsatellite SMM and GSM mutation models, locus-specific mutation parameters, and composite parameters including θ = 4Neµ and τ = tµ.The same mutation model must be applied to all loci.
- ABC analyses: DIYABC can build or extend reference tables, compare scenario posterior probabilities, estimate posterior parameter distributions, assess bias and precision, and simulate Genepop-format data.Available summary statistics include mean allele number, genic diversity, Fst, (δµ)2, and admixture rates.
- Simulated example: In a simulated six-population example with three splits and two admixture events, both direct and logistic-regression approaches significantly favored scenario 1 over two simpler alternatives.The analysis used 60,000 closest simulations for logistic regression and 600 for the direct approach.
- Simulated example: Posterior estimates were informative for common effective population size and admixture rates but not for all event times, while confidence testing found substantial errors for scenario 1.Using 500 test data sets, type I errors for scenario 1 were 0.414 with the direct approach and 0.3 with logistic regression.
4 CONCLUSION
DIY ABC makes ABC-based inference accessible to non-specialists for complex population evolutionary scenarios, while supporting scenario comparison and confidence assessment. Its current limitations are the assumed absence of post-divergence migration and mutation models focused mainly on microsatellites.
- DIY ABC enables non-specialists to perform ABC-based inference on complex population evolutionary scenarios without reducing them to standard cases.
- The program compares competing scenarios and quantifies their relative support from the data.
- DIY ABC provides a way to evaluate confidence in its parameter estimates.
- The current version assumes no migration among populations after divergence and relies mostly on microsatellite mutation models.Future developments were intended to remove these limitations progressively.
FUNDING
The development of DIY ABC was supported by French and European research grants, including funding for a fellowship at Imperial College London.
- DIY ABC received support from a French National Research Agency grant and an EU Marie-Curie Fellowship grant.The fellowship supported two years in Imperial College London’s Epidemiology and Public Health department, where a major part of the program was written.