Stages of a genome-wide association study

(1, 2) Well-characterized cases and controls of similar ancestry are genotyped using oligonucleotide microarrays, typically consisting of assays for 500,000 to 1,000,000 SNPs. A representative cartesian plot of results for one such assay is presented. Individuals homozygous for the major allele are plotted in blue, homozygotes for the minor allele in red, and heterozygotes in green.
(3) Next, extensive quality control analysis is conducted, including assessments at the level of the individual samples (microarray quality, subject genotype completion rates, gender inconsistencies) and individual assays (assay quality scores, genotype completion rates, Hardy Weinberg equilibrium, parent-child genotype inconstancies). Screens for cryptic sample relatedness are performed using Identity By State statistics. Data are then filtered, with removal of results for problematic subjects or assays.
(4) Principal components analysis (shown) enables screening for evidence of population stratification. If detected, outlier subjects can be removed from consideration, and the eigenvectors generated can be used as covariates in the subsequent association testing.
(5) Using reference genotype panels, additional SNP genotypes for untyped variants (grey SNPs) can be imputed from the typed variants (in black) and considered for association testing.
(6) Statistical inference is performed by testing individual variants for association by logistic regression for case-control studies or linear regression for quantitative traits, with adjustment for relevant covariates, including eigenvectors. To accommodate the large number of hypotheses tested, p-values must be adjusted, with more stringent cutoffs p-values designated in advance.
(7) Results can be visualized using so-called "Manhattan Plots," where the individual -log10 p-values are plotted as a function of physical position.
(8) Significant associations can be passed for additional validation, including replication studies in independent cohorts, sequencing and fine-mapping studies to identify causal variants, and experimental validation to confirm biological relevance and function. Additional modeling, including gene-by-gene interaction and gene-by-environment modeling, can also be conducted.