A simulation-based evaluation of methods for inferring linear barriers to gene flow

Area of Science:

• Population genetics
• Bioinformatics
• Statistical ecology

Background:

• Analytical techniques can yield differing conclusions on genetic structure.
• Reliable interpretation necessitates evaluating the efficacy of statistical methods.

Purpose of the Study:

• To evaluate multiple analytical methods for detecting linear barriers to gene flow.
• To assess the impact of simulation conditions (dispersal, genetic equilibrium) on barrier detection power.

Main Methods:

• Evaluated boundary detection (Monmonier's algorithm, WOMBLING), spatial Bayesian clustering (TESS, GENELAND), aspatial clustering (STRUCTURE), and non-Bayesian clustering (PSMIX, DAPC).
• Simulated data under varying dispersal abilities and genetic equilibrium to test method performance.

Main Results:

• Clustering methods demonstrated higher success rates and faster barrier detection compared to boundary detection methods.
• Long-distance dispersal accelerated barrier detection across all methods.
• Bayesian clustering methods, especially GENELAND, exhibited superior performance in success rate and detection speed.
• No methods detected a continuous linear barrier under isolation-by-distance (IBD) models, though clustering methods risked incorrect inferences without strict criteria.

Conclusions:

• Bayesian clustering methods are recommended for detecting linear barriers to gene flow.
• Method selection and interpretation require careful consideration of simulation conditions, particularly dispersal patterns and potential for isolation-by-distance.
• GENELAND shows high power for barrier detection, but all clustering methods need cautious application to avoid false positives under IBD.