Detecting the number of clusters of individuals using the software STRUCTURE: a simulation study

Area of Science:

• Population genetics
• Computational biology
• Bioinformatics

Background:

• Identifying genetically homogeneous groups is a fundamental challenge in population genetics.
• The Bayesian algorithm in STRUCTURE software is widely used for this purpose.
• Its effectiveness with non-homogeneous dispersal patterns remains untested.

Purpose of the Study:

• To evaluate the STRUCTURE algorithm's ability to detect the correct number of genetic clusters (K).
• To test this under various non-homogeneous population dispersal scenarios.
• To assess the utility of the 'log probability of data' and a novel DeltaK statistic.

Main Methods:

• Utilized an individual-based model to generate genetic data under diverse dispersal scenarios.
• Applied the STRUCTURE Bayesian algorithm to analyze simulated datasets.
• Calculated the 'log probability of data' and the ad hoc DeltaK statistic for successive K values.

Main Results:

• The standard 'log probability of data' often failed to accurately estimate the true number of clusters (K).
• The DeltaK statistic effectively identified the uppermost hierarchical level of genetic structure across tested scenarios.
• Results demonstrated sensitivity to marker type (AFLP vs. microsatellite), loci number, populations sampled, and individuals typed.

Conclusions:

• The DeltaK statistic offers a more reliable method for inferring the number of genetic clusters with STRUCTURE.
• Accurate detection of population structure is feasible even with complex dispersal patterns.
• Optimal results depend on careful consideration of genetic marker choice and data sample size.