Neural posterior estimation for population genetics.

Area of Science:

• Population Genetics
• Computational Biology
• Machine Learning

Background:

• Simulation-based inference methods, like Approximate Bayesian Computation (ABC), are valuable in population genetics but face computational expense and limitations with high-dimensional data.
• Supervised machine learning (ML) offers an alternative but typically lacks Bayesian uncertainty estimates.

Purpose of the Study:

• To introduce and evaluate Neural Posterior Estimation (NPE) as a method combining the strengths of ABC and supervised ML for population genetics.
• To demonstrate NPE's accuracy, efficiency, and applicability in demographic inference using genetic data.

Main Methods:

• Trained a neural network to perform Neural Posterior Estimation (NPE) for population genetics models.
• Compared NPE with existing inference methods using raw genotypes and summary statistics as input.
• Applied NPE to demographic inference for both simple and complex population models.

Main Results:

• Neural posterior estimators demonstrated high accuracy and efficiency in yielding posterior distributions.
• NPE successfully estimated posterior distributions using both raw genetic data and summary statistics.
• The method proved effective for demographic inference in various population genetic scenarios.

Conclusions:

• Neural Posterior Estimation (NPE) provides a powerful and versatile approach for complex population genetics inference.
• NPE overcomes key limitations of Approximate Bayesian Computation (ABC) and traditional machine learning.
• A user-friendly workflow is provided to facilitate the adoption of NPE in population genetics research.