PLINK

PLINK provides high-speed utilities for manipulating large-scale genetic datasets in binary (BED/BIM/FAM) and text (PED/MAP) formats. It efficiently handles variant subsetting, sample filtering, allele recoding, and strand flipping to harmonize data from different genotyping platforms. The --recode and --make-bed commands are fundamental for converting between formats and preparing data for downstream PRS software like PRSice or LDpred2.

PLINK implements rigorous per-variant and per-sample quality control filters essential for removing technical artifacts before GWAS or PRS computation. Key filters include:

• Missingness: --geno and --mind remove variants and samples with high call rates.
• Minor Allele Frequency (MAF): --maf excludes rare variants with low statistical power.
• Hardy-Weinberg Equilibrium (HWE): --hwe flags genotyping errors in controls.
• Sex Check: --check-sex identifies sample mismatches using X-chromosome heterozygosity.

PLINK's --indep-pairwise command performs LD-based variant pruning, a critical step for the Clumping and Thresholding (C+T) method of PRS construction. It slides a window across the genome, calculating the squared correlation (r²) between variants, and removes those in high LD to create a set of approximately independent index SNPs. This prevents double-counting of genetic effects from correlated variants and is essential for calculating principal components for population stratification correction.

PLINK performs efficient genome-wide association studies using linear or logistic regression under an additive genetic model. The --assoc command runs a basic allelic test, while --linear and --logistic incorporate covariates like age, sex, and principal components to control for confounding. These commands generate the summary statistics—beta coefficients, standard errors, and p-values—that serve as the base data for downstream PRS methods like PRS-CS and LDpred2.

PLINK integrates Principal Component Analysis (PCA) via --pca to quantify and correct for population structure, a major confounder in GWAS and PRS analyses. The resulting eigenvectors can be included as covariates in association models to prevent spurious genotype-phenotype correlations. Additionally, --genome estimates pairwise identity-by-descent (IBD) to identify cryptic relatedness, and --cluster performs identity-by-state (IBS) clustering for homogenous cohort selection.

PLINK's --score command directly computes a weighted polygenic risk score by summing the product of the effect allele dosage and the effect size (beta) across a set of variants. It takes a file of variant IDs, effect alleles, and weights (from a GWAS) and outputs a per-individual score. While basic, this function is the direct implementation of the Clumping and Thresholding method and is widely used for rapid prototyping and validation of PRS models before deploying more sophisticated Bayesian methods.

A hypothesis-free statistical framework that scans millions of single nucleotide polymorphisms across the genome to identify genotype-phenotype associations. PLINK serves as the primary computational engine for executing GWAS on large population cohorts.

• Uses linear or logistic regression for quantitative and binary traits
• Outputs summary statistics including beta coefficients, standard errors, and p-values
• Requires rigorous multiple testing correction (Bonferroni threshold typically p < 5×10⁻⁸)

A critical quality control step that removes correlated genetic variants to ensure independence among predictors. PLINK's --indep-pairwise command implements a sliding window approach to retain only tag SNPs.

• Parameters: window size (e.g., 50 variants), step size (5), and r² threshold (0.2)
• Essential before principal component analysis and clumping and thresholding
• Reduces computational burden and eliminates redundant association signals

A dimensionality reduction technique implemented in PLINK via --pca to infer continuous axes of genetic ancestry. The resulting eigenvectors serve as covariates in association models to correct for population stratification.

• Typically retains the top 10–20 principal components
• Visualized by plotting PC1 vs. PC2 to identify population outliers
• Prevents spurious associations driven by systematic ancestry differences between cases and controls

The foundational method for constructing polygenic risk scores directly from GWAS summary statistics. PLINK's --clump command selects independent index variants, while --score computes the weighted sum of risk alleles.

• Clumping: retains the most significant variant within each LD block
• Thresholding: includes only variants below a specified p-value cutoff (e.g., p < 5×10⁻⁸, p < 0.05)
• Produces a single per-individual PRS that can be evaluated for predictive performance

PLINK provides a comprehensive suite of per-variant and per-sample filters to remove genotyping errors and low-quality data before association testing. Standard thresholds ensure robust, reproducible results.

• Variant-level: minor allele frequency (--maf 0.01), genotype missingness (--geno 0.05), Hardy-Weinberg equilibrium (--hwe 1e-6)
• Sample-level: individual missingness (--mind 0.1), heterozygosity outliers, cryptic relatedness (--genome)
• Sex check: identifies discrepancies between reported and genetically inferred sex using X-chromosome homozygosity

A Bayesian PRS method that supersedes simple clumping and thresholding by modeling the full genetic architecture. Unlike PLINK's C+T approach, LDpred2 uses a point-normal mixture prior on variant effect sizes and a Gibbs sampler to infer posterior mean effects.

• Requires GWAS summary statistics and an external LD reference panel
• Accounts for LD patterns rather than simply pruning correlated variants
• Often achieves higher variance explained (R²) than C+T in out-of-sample prediction