Complex genetic admixture histories reconstructed with Approximate Bayesian Computation

Area of Science:

• Population Genetics
• Evolutionary Biology
• Computational Biology

Background:

• Admixture is a key evolutionary process shaping genetic diversity across species.
• Inferring complex admixture histories, especially beyond simple pulse models, is methodologically challenging using traditional maximum-likelihood approaches.
• Existing methods struggle with intricate admixture scenarios involving multiple or recurring admixture events.

Purpose of the Study:

• To develop an Approximate Bayesian Computation (ABC) framework for reconstructing complex admixture histories using independent genetic markers.
• To create a simulation software package, MetHis, capable of modeling diverse admixture scenarios.
• To apply this novel framework to analyze the admixture history of specific human populations.

Main Methods:

• Developed the MetHis software package to simulate genetic data under various complex admixture scenarios (multiple pulses, recurring admixture).
• Utilized independent genetic markers (SNPs, microsatellites) and user-defined prior distributions for simulation parameters.
• Coupled MetHis with machine-learning ABC algorithms (random forest for scenario choice, neural networks for parameter estimation) to analyze simulated and empirical data.

Main Results:

• Random forest ABC accurately distinguished between most complex admixture scenarios, with errors primarily in biologically similar, nested parameter spaces.
• Neural network ABC provided accurate and conservative posterior parameter estimation for complex admixture scenarios.
• Analysis of African American and Barbadian populations indicated that monotonically decreasing ancestral contributions over time provided a better fit than multiple admixture pulses.

Conclusions:

• The developed ABC framework and MetHis software offer a powerful tool for reconstructing detailed and complex admixture histories.
• This approach overcomes limitations of maximum-likelihood methods for intricate demographic scenarios.
• The findings suggest that decreasing admixture over time is a more accurate model for the demographic history of the studied populations.