Overview
Genome-wide association studies (GWAS) are large-scale analyses that scan the genomes of thousands of individuals to identify genetic variants associated with particular traits or diseases. By comparing allele frequencies between cases and controls, GWAS can pinpoint genomic loci that contribute to complex phenotypes such as height, diabetes, or cardiovascular disease. The approach is hypothesis-free, it examines millions of SNPs across the genome without prior assumptions about which genes are involved. Since the first landmark GWAS in 2005, this method has uncovered tens of thousands of trait-associated loci.
Methods
A typical GWAS proceeds through several stages. First, study participants are genotyped using SNP arrays or DNA sequencing. After quality control (filtering SNPs by call rate, Hardy-Weinberg equilibrium, and minor allele frequency), the data undergo imputation to infer untyped variants using reference panels. Association testing is performed using logistic regression or linear mixed models that correct for population stratification. The stringent genome-wide significance threshold (p < 5 × 10⁻⁸) accounts for multiple testing across millions of SNPs. Results are visualized in Manhattan plots showing chromosomal positions versus significance, and Q-Q plots assess systematic bias.
Applications
GWAS has transformed our understanding of common disease genetics. It has identified hundreds of loci for type 2 diabetes, coronary artery disease, and autoimmune disorders, many pointing to unexpected biological pathways. In cancer research, GWAS loci have revealed new susceptibility genes and potential cancer biochemistry targets. However, translating GWAS signals into causal mechanisms remains challenging, as most associated variants lie in non-coding regions. Post-GWAS analyses include fine-mapping, functional annotation, and Mendelian randomization to establish causality and explore clinical biochemistry relevance.
Practical Protocol
A GWAS pipeline begins with genotype data from SNP arrays or imputed sequencing data. Sample-level QC filters out individuals with call rate <95%, heterozygosity outliers (±3 SD from mean), and sex mismatches. SNP-level QC removes variants with call rate <95%, Hardy-Weinberg equilibrium p < 1×10⁻⁶ (in controls), and minor allele frequency (MAF) < 1%. All QC is performed in PLINK: plink --bfile data --geno 0.05 --hwe 1e-6 --maf 0.01 --mind 0.05 --make-bed --out data_qc. Imputation to a reference panel (e.g., 1000 Genomes or TOPMed) uses a two-step process: pre-phasing with SHAPEIT4 or Eagle2, then imputation with Minimac4 or IMPUTE5 on a cloud server like the Michigan Imputation Server. After imputation, convert dosages to hard calls and filter by Rsq > 0.3. Population stratification is assessed with principal component analysis: plink --bfile data_qc --pca 10 --out pca_results. Association testing uses a linear mixed model implemented in SAIGE or BOLT-LMM for binary or quantitative traits, correcting for relatedness and population structure: saige --vcf imputed.vcf --traitType binary --pheno phenotype.txt --covar pca_results.eigenvec --output assoc.txt. The genome-wide significance threshold is p < 5×10⁻⁸. Visualize results with a Manhattan plot using the qqman R package: manhattan(gwas, chr="CHR", bp="BP", p="P", snp="SNP", suggestiveline = -log10(1e-5), genomewideline = -log10(5e-8)). Assess inflation with the genomic inflation factor λ; values <1.05 indicate minimal stratification. Fine-map significant loci with FINEMAP or SuSiE to identify credible sets of causal variants within associated regions.
