Two Adjustment Strategies for Imputation across Genotyping Arrays

• Published in: Human heredity Vol. 78; no. 2; pp. 73 - 80
• Main Authors: Xie, Yiran, Hancock, Dana B., Johnson, Eric O., Rice, John P.
• Format: Journal Article
• Language: English
• Published: Basel, Switzerland S. Karger AG 01.01.2014
• Subjects: Case-Control Studies; False Positive Reactions; Genotype; Genotype & phenotype; Genotyping Techniques - methods; Genotyping Techniques - statistics & numerical data; Humans; Original Paper; Polymorphism; Polymorphism, Single Nucleotide; Tobacco Use Disorder - genetics; Genome-wide association studies; Imputation; Genotyping arrays; Spurious associations
• ISSN: 0001-5652; 1423-0062; 1423-0062
• DOI: 10.1159/000363337

Genotype imputation is a powerful approach in genomewide association studies (GWAS) because it can provide higher resolution for associated regions and facilitate metaanalysis. However, bias can exist if different genotyping arrays are used and are unbalanced for case versus control subjects. The intersection imputation strategy [imputation based on single nucleotide polymorphisms (SNPs) available on all arrays] is a valid strategy that eliminates the bias caused by unbalanced genotyping, but achieved at the expense of reduced statistical power. In order to improve power in this situation, we introduce two new strategies: the replacement strategy based on the imputation quality score (IQS0) ≥ 0.9 and the correction strategy. The IQS is a score that we have previously introduced based on Cohen’s kappa of rater agreement. The replacement strategy with IQS ≥ 0.9 is a hybrid approach that utilizes measured genotypes for SNPs available on one or more of all arrays whenever the SNP has a high imputation quality (defined by IQS ≥ 0.9). The correction strategy combines measured genotypes as well as imputed and corrected genotype dosages for SNPs available on one or more of all arrays. The correction strategy yields a valid statistical test, while the replacement strategy with IQS ≥ 0.9 eliminates most spurious associations. Both strategies maintain statistical power.