In this paper, we are concerned with problems occurring one step further down the pipeline. Under the assumption that markers actually genotyped by each chip are being genotyped with good accuracy, we investigate how well Type I error rates are maintained after imputation in a study where cases and controls are genotyped on different platforms. To do this, we used the healthy control groups from two studies nested within the Nurses’ Health Study: a Type 2 Diabetes (T2D) study genotyped on Affy, and a Breast Cancer (BrCa) study genotyped on Illumina. After imputation within each study, we label the T2D controls “cases” and the BrCa controls “controls,” and fit a logistic regression predicting this case-control status from each SNP. We expect there to be no substantial genetic differences between these two groups –so any significant differences we see reflect a Type I error rate higher than expected.
