Reliably discriminating stock structure with genetic markers: Mixture models with robust and fast computation

Area of Science:

• Population genetics
• Fisheries science
• Computational biology

Background:

• Accurate delineation of fish stocks is vital for effective natural resource management and sustainable harvesting.
• Current stock delineation methods often inappropriately borrow from human genetics, focusing on ancestry (admixture) rather than breeding populations.

Purpose of the Study:

• To evaluate the suitability of ancestral analysis methods for stock delineation in wild species.
• To propose and validate a direct stock identification model and robust computational methods for this purpose.

Main Methods:

• Simulation experiments comparing ancestral analysis with a direct stock identification model.
• Analysis of yellowfin tuna data using proposed and established methods.
• Development and comparison of alternative computational strategies, including an expectation-maximization algorithm.

Main Results:

• Ancestral analysis methods are not always appropriate for stock delineation, potentially yielding misleading point estimates.
• A simpler, direct stock identification model is advocated.
• A robust expectation-maximization algorithm offers a computationally efficient and accurate approach to stock delineation and uncertainty quantification.

Conclusions:

• Direct stock identification models are more appropriate than ancestral analysis for delineating wild fish populations.
• Robust computational methods, like the proposed expectation-maximization algorithm, are essential for accurate and efficient stock delineation.
• Quantifying uncertainty in model parameters and assignment probabilities is critical to avoid misleading conclusions.