Accelerated Bayesian inference of population size history from recombining sequence data

Area of Science:

• Population genetics
• Computational biology
• Genomic data analysis

Background:

• Inferring population history is crucial for understanding evolutionary processes.
• Existing methods for population history inference from whole genome data have limitations in speed and accuracy.

Purpose of the Study:

• To introduce PHLASH, a novel Bayesian method for accurate and efficient population history inference.
• To provide a robust tool for analyzing whole genome sequence data and quantifying uncertainty.

Main Methods:

• PHLASH employs Bayesian inference by averaging sampled histories derived from low-dimensional projections of the coalescent intensity function.
• It utilizes a novel algorithm for computing the score function of coalescent hidden Markov models, optimized for GPU hardware.
• The method integrates a PSMC-like model for size history estimation.

Main Results:

• PHLASH demonstrates superior speed and lower error rates compared to SMC++, MSMC2, and FITCOAL on simulated data.
• The method provides a full posterior distribution, enabling automatic uncertainty quantification.
• New Bayesian testing procedures for population structure and ancient bottlenecks are enabled.

Conclusions:

• PHLASH offers an accurate, efficient, and user-friendly Python package for population history inference.
• Its novel algorithmic approach and GPU optimization advance the field of computational population genetics.