Full likelihood inference from the site frequency spectrum based on the optimal tree resolution.

Area of Science:

• Population Genetics
• Computational Evolutionary Biology
• Statistical Inference

Background:

• Accurate inference of population genetic scenarios requires computing likelihood functions based on observed genetic data, such as the site frequency spectrum (SFS).
• Traditional methods can be computationally intensive, especially when dealing with complex demographic histories and population structures.
• Existing approaches often struggle with the high dimensionality of the genealogical state space.

Purpose of the Study:

• To develop a novel importance sampler for efficiently computing the full likelihood function of demographic and structural scenarios.
• To reduce the computational burden by utilizing only the necessary information from genealogical trees for SFS likelihood calculation.
• To accommodate models of demographic change and non-panmictic population structure.

Main Methods:

• Developed a novel importance sampler that represents genealogies using minimal information required for SFS likelihood.
• Employed a controlled Markov process to generate compatible SFS histories ('particles').
• Utilized Aldous' Beta-splitting model for prior distributions on genealogical topologies and parametric priors for demographic models (e.g., exponential growth, bottlenecks).

Main Results:

• The importance sampler significantly reduces the state space complexity for likelihood computation.
• The method effectively integrates demographic changes and population structure into the likelihood framework.
• Demonstrated capability to estimate population scenario likelihoods using independent SFS data and distinguish between different population models.

Conclusions:

• The novel importance sampler provides an efficient and flexible tool for population genetic inference.
• This approach enhances the ability to model complex demographic histories and population structures from SFS data.
• The method offers a significant advancement in computational population genetics, enabling more robust evolutionary analyses.