'Highly-Informative' Genetic Markers Can Bias Conclusions: Examples and General Solutions

Area of Science:

• Population Genetics
• Genomic Data Analysis
• Conservation Science

Background:

• High-grading bias arises from model overfitting, leading to overestimated power in selected genetic loci.
• This bias occurs when highly informative or high-F_ST markers are chosen and reused for subsequent analyses, a common practice in population genetics.
• The issue can arise even in panmictic populations without local adaptation, potentially impacting conservation strategies.

Purpose of the Study:

• To demonstrate how high-grading bias can severely overestimate population structure.
• To highlight the potential downstream implications for conservation, including erroneous delineation of conservation units.
• To introduce methods for detecting and preventing high-grading bias in genetic datasets.

Main Methods:

• Analysis of empirical and simulated datasets to identify high-grading bias.
• Evaluation of marker ascertainment strategies, including F_ST-based selection.
• Demonstration of alternative methods like statistically based outlier tests, permutation tests, and cross-evaluation for bias detection.

Main Results:

• High-grading bias can lead to significant overestimation of population structure, even in the absence of true differentiation.
• The practice of selecting high-F_ST markers can result in spurious population structure, impacting conservation decisions.
• Statistically based outlier tests, permutation tests, and cross-evaluation were shown to be effective in reducing or detecting high-grading bias.

Conclusions:

• High-grading bias is a critical concern in population genetic studies, potentially leading to flawed conservation actions.
• Researchers should be cautious when using subsets of markers, especially those selected for high differentiation.
• The R package PCAssess is provided to assist researchers in detecting and mitigating high-grading bias through automated analyses.