Population Structure (PCA & ADMIXTURE)

How it works

Principal Component Analysis (PCA) on the genotype matrix gives a fast, model-free view of population stratification. The top PCs are essential covariates in GWAS — without them, subpopulation differences inflate false positives. We complement PCA with an ADMIXTURE-style ancestry analysis that estimates fractional ancestry from K subpopulations and uses cross-validation error to choose the best K.

Formula

What you get

• ▸PC1–PC4 scatter plots with percent variance explained
• ▸Ancestry-proportion stacked bars for K=2..8
• ▸Cross-validation error curve to select K

When to use it

• ▸Before any GWAS run on a diverse panel
• ▸When sampling parents for a breeding program from multiple gene pools
• ▸To verify dataset composition before genomic selection

Inputs

Parameters

Workflow

1. 1. Prune
   Sliding-window LD pruning to independent markers.
2. 2. PCA
   Eigendecomposition of the standardised marker matrix.
3. 3. ADMIXTURE grid
   Fit K = 2..8, compute cross-validation error at each K.
4. 4. Pick K
   Elbow of CV error; visualise stacked ancestry bars sorted by K=best.

Interpreting results

• ▸Clear cluster separation on PC1/PC2 = strong subpopulation structure; use ≥3 PCs downstream.
• ▸Flat ADMIXTURE bars with fractions near 0.5 across all K = admixed material.
• ▸PC-variance-explained percentages are a quick check on how much of the panel's diversity is structural.

Common pitfalls

• ✕Skipping LD pruning makes PC1 track a single high-LD region rather than genome-wide structure.
• ✕Reporting a single 'true' K is over-confident — always show the CV curve.
• ✕Related individuals (sibs) collapse to the same PC point and bias K selection.

Worked example

Try it — interactive example

References

• Patterson et al. (2006) PLoS Genetics — Population structure and eigenanalysis
• Alexander, Novembre & Lange (2009) Genome Research — ADMIXTURE