The computer program structure for assigning individuals to populations: easy to use but easier to misuse

Area of Science:

• Population genetics
• Computational biology
• Evolutionary biology

Background:

• The Structure program uses Bayesian methods for population genetics analysis.
• It is widely applied in ecology, evolutionary biology, human genetics, and conservation biology.
• Previous studies suggest Structure yields erroneous inferences with unbalanced sample sizes.

Purpose of the Study:

• To confirm Structure's poor performance with unbalanced sampling.
• To investigate the impact of ancestry priors on Structure's accuracy.
• To identify optimal parameter settings for accurate inferences in challenging scenarios.

Main Methods:

• Analysis of simulated and empirical population genetics data.
• Comparison of Structure's performance using default versus alternative ancestry priors.
• Evaluation of different parameter combinations, including ALPHA values and allele frequency models.

Main Results:

• This study confirms Structure yields poor individual assignments and incorrect population number (K) estimates with unbalanced sampling.
• The default ancestry prior is identified as the primary cause of poor performance.
• Adopting an alternative ancestry prior significantly improves assignment accuracy, even with highly unbalanced samples.
• The alternative prior also enhances K inference, though to a lesser extent than individual assignments.
• Specific parameter combinations (alternative prior, low ALPHA, uncorrelated allele frequency model) are crucial for accurate results in complex cases.

Conclusions:

• Structure's performance is highly sensitive to sampling balance and prior settings.
• The default ancestry prior can lead to significant errors; the alternative prior offers a robust solution.
• Careful selection of parameters, especially for complex scenarios, is essential for reliable Structure analysis.
• Users should employ multiple models and estimators for robust and exploratory analyses.