Distinguishing among complex evolutionary models using unphased whole-genome data through random forest approximate...

Area of Science:

• Population genetics
• Computational biology
• Genomics

Background:

• Inferring past demographic histories is crucial for understanding population genetics.
• Current inferential methods face computational limitations with large genomic datasets.
• Evaluating the reliability of estimates further increases computational effort.

Purpose of the Study:

• To develop an efficient computational framework for inferring complex demographic histories using complete genomes.
• To introduce a novel statistic, the full genomic distribution of the four mutually exclusive categories of segregating sites (FDSS), for summarizing genomic data.
• To assess the accuracy and efficiency of the proposed framework in recognizing true demographic models.

Main Methods:

• Developed an approximate Bayesian computation framework based on the random forest algorithm (ABC-RF).
• Utilized the FDSS statistic, which is fast to compute from unphased genome data and does not require ancestral allele states.
• Tested the framework using simulated data with varying sampling strategies and compared FDSS with site frequency spectrum (SFS) statistics.

Main Results:

• The ABC-RF framework accurately recognizes true demographic models among increasingly complex scenarios.
• The FDSS statistic demonstrates effectiveness in summarizing genomic data for demographic inference.
• The study identified optimal experimental conditions for maximizing inferential power using ABC-RF and FDSS.

Conclusions:

• The proposed ABC-RF framework offers an efficient solution for inferring complex demographic histories from complete genomes.
• FDSS provides a valuable and computationally efficient statistic for population genetic analyses.
• The framework was successfully applied to real-world datasets, including human dispersal out of Africa and orangutan evolutionary relationships.