Q&A 9 How do you perform PCA on genotype data to assess population structure?

9.1 Explanation

PCA (Principal Component Analysis) is a standard method to explore genetic diversity and detect hidden population structure in GWAS datasets. It helps:

• Control for confounding due to stratification
  
• Visualize sample clusters or outliers
  
• Generate covariates (PC1, PC2, etc.) for use in association models

We apply PCA on the imputed genotype dosage matrix, excluding identifier columns (FID, IID), and then combine the results with sample IDs.

9.2 R Code

# Load libraries
library(tidyverse)

# Step 1: Extract genotype matrix (exclude FID and IID)
geno_numeric <- geno_imputed[, -c(1, 2)]

# Step 2: Perform PCA using prcomp
pca_result <- prcomp(geno_numeric, center = TRUE, scale. = TRUE)

# Step 3: Combine first 5 PCs with sample IDs
pca_df <- geno_imputed[, 1:2] %>%  # FID and IID
  bind_cols(as_tibble(pca_result$x[, 1:5]))  # PC1 to PC5

# Step 4: Plot PC1 vs PC2
ggplot(pca_df, aes(x = PC1, y = PC2)) +
  geom_point(size = 2, alpha = 0.7) +
  labs(title = "PCA of Genotype Data", x = "PC1", y = "PC2") +
  theme_minimal()

✅ Takeaway: PCA helps uncover hidden structure in your GWAS population. Always reattach FID and IID to PCA scores so they can be merged with phenotype and genotype metadata for downstream modeling.