Extending approximate Bayesian computation with supervised machine learning to infer demographic history from genetic...

Area of Science:

• Population Genetics
• Computational Biology
• Bioinformatics

Background:

• Simulation-based methods like approximate Bayesian computation (ABC) are crucial for analyzing complex population and species genetic histories.
• Supervised machine learning (SML) offers efficient statistical solutions for scenario choice and parameter estimation in population genetics.

Purpose of the Study:

• To implement Random Forest (RF) algorithms for efficient inferences in population genetics.
• To develop a user-friendly package (DIYABC Random Forest v1.0) integrating simulation and RF analysis for molecular data.
• To evaluate the accuracy and power of RF methods for inferring complex population genetic histories from various data types, including large SNP datasets.

Main Methods:

• Implemented Random Forest (RF) algorithms using simulated datasets from DIYABC v2.1.0.
• Developed DIYABC Random Forest v1.0, a package for simulating molecular data (microsatellites, DNA sequences, SNPs) and performing RF analyses.
• Applied RF methods for scenario choice and parameter estimation on pseudo-observed and real SNP datasets (pool-sequencing and individual-sequencing).

Main Results:

• DIYABC Random Forest v1.0 provides a user-friendly interface for genetic data simulation and RF-based inference.
• RF methods enable efficient inferences at low computational cost, bypassing the need for preliminary summary statistic selection and ABC tolerance level derivation.
• The package demonstrates effectiveness in analyzing large SNP datasets for complex population genetic history inferences.

Conclusions:

• DIYABC Random Forest v1.0 offers a powerful and efficient tool for population geneticists.
• The integration of RF methods significantly enhances the analysis of complex genetic datasets, particularly large SNP datasets.
• This approach facilitates robust scenario choice and parameter estimation in evolutionary studies.