Artifact due to differential error when cases and controls are imputed from different platforms.

Area of Science:

• Genetics
• Bioinformatics
• Statistical genomics

Background:

• Genome-wide association studies (GWAS) are powerful tools for identifying genetic variants associated with diseases.
• Combining samples genotyped on different platforms can lead to cost savings but introduces design biases.
• Imputation of genotypes across platforms can result in differential measurement error and inflated Type I error rates (false positives).

Purpose of the Study:

• To evaluate the impact of combining control samples genotyped on different platforms in GWAS.
• To assess methods for controlling Type I error inflation caused by platform differences and imputation.
• To provide recommendations for study design and analysis when merging data from disparate genotyping platforms.

Main Methods:

• Comparison of genotype frequencies between two control groups from the Nurses' Health Study genotyped on Affymetrix 6.0 and Illumina HumanHap550 platforms.
• Application of standard imputation quality filters to identify significant single-nucleotide polymorphisms (SNPs) at a genome-wide significance level (5 × 10(-8)).
• Evaluation of three methods to control Type I error: principal components for platform effects, restricting to high-quality imputation SNPs, and re-genotyping controls.

Main Results:

• A significant number of SNPs (9,841 out of 2,347,809, 0.4%) met genome-wide significance thresholds due to platform differences.
• Principal component analysis for platform effects did not reduce the Type I error rate.
• Restricting to high-quality imputation SNPs and re-genotyping controls significantly reduced the error rate but required excluding some SNPs.

Conclusions:

• Combining control samples genotyped on different platforms without careful consideration can lead to substantial inflation of Type I error rates in GWAS.
• Methods such as restricting to high-quality imputation data or re-genotyping controls can mitigate these false positives.
• Researchers should prioritize eliminating biases at the design stage by genotyping sufficient samples on each platform and validate all genome-wide significant findings on an independent platform.