DIYABC

DIYABC Random Forest v1.0 applies Random Forest–based Approximate Bayesian Computation (ABC) to perform model choice and parameter inference in population genetics using simulated datasets for microsatellites, DNA sequences, and single nucleotide polymorphisms (SNPs).

Key Features:

• ABC–RF integration: Integrates Approximate Bayesian Computation (ABC) with supervised Random Forests (RF) to perform inference without preliminary selection of ABC summary statistics or derivation of tolerance levels.
• Simulation engine: Generates simulated datasets using an enhanced population genetic simulator from DIYABC v2.1.0 for custom evolutionary scenarios.
• Supported data types: Supports microsatellites, DNA sequences, and SNPs, including pool-sequencing and individual-sequencing SNP data.
• Feature vector: Utilizes an extensive feature vector comprising various summary statistics and their linear combinations.
• RF-based statistical treatments: Implements RF algorithms for model choice and parameter inference and includes statistical tools to evaluate power and accuracy.
• Scalability: Capable of handling large SNP datasets.

Scientific Applications:

• Model choice (scenario choice): Distinguishes among competing evolutionary scenarios using ABC coupled with Random Forests.
• Parameter inference: Estimates demographic and evolutionary parameters from simulated and observed genetic data.
• SNP data analysis: Applies to both pool-sequencing and individual-sequencing SNP datasets for inference tasks.
• Performance assessment: Evaluates power and accuracy of inferences using RF-based statistical tools and pseudo-observed datasets.

Methodology:

Simulations are produced with the enhanced DIYABC v2.1.0 population genetic simulator across microsatellites, DNA sequences, and SNPs; Random Forest algorithms process simulated datasets using feature vectors of summary statistics and their linear combinations to perform model choice and parameter inference, with RF-based statistical evaluations of power and accuracy.