Estimation of demo-genetic model probabilities with Approximate Bayesian Computation using linear discriminant...

Area of Science:

• Population Genetics
• Computational Biology
• Statistical Ecology

Background:

• Approximate Bayesian Computation (ABC) is widely used for comparing demographic models with molecular data.
• Large numbers of populations, models, and summary statistics present significant computational challenges in ABC.
• Previous work highlighted the need for additional simulations to validate ABC model comparison conclusions.

Purpose of the Study:

• To develop a computationally efficient method for Approximate Bayesian Computation (ABC) scenario probability calculation.
• To assess the precision and speed of a novel method using linear discriminant analysis (LDA) on summary statistics.
• To evaluate the practical application of this method for inferring invasion routes using real genetic data.

Main Methods:

• Linear Discriminant Analysis (LDA) applied to summary statistics (Ss) prior to logistic regression for scenario probability computation.
• Comparison of LDA-transformed Ss with raw Ss using simulated pseudo-observed data sets (pods).
• Assessment of precision, Type I and II errors, and computation time.

Main Results:

• Scenario probabilities derived from LDA-transformed Ss strongly correlate with those from raw Ss.
• The LDA method demonstrated comparable precision and error rates to traditional methods.
• A significant speed gain (approximately 100-fold) was achieved using LDA-transformed Ss, greatly enhancing computational efficiency.

Conclusions:

• The proposed LDA-based method provides a computationally efficient and accurate approach for ABC model comparison.
• This innovation enables the analysis of a larger number of pseudo-observed data sets and complex scenarios.
• The method offers a manageable way to empirically evaluate the power of discrimination among numerous complex demographic models in population genetics.