In the two Illumina projects, we did not detect unusual patterns of relatedness or evidence of mixed samples. No low quality samples were detected in the QA process for the Lung Cancer project. In the Addiction project, one problematic genotyping batch of 3 samples was detected and removed from the data set (of which only one sample was used in cluster definition). For Illumina genotype calling, all samples (except those with call rate < 98%) are used for cluster definition, so this system is much less susceptible to the influence of a few low quality samples than the by-plate Birdseed calling for Affymetrix projects. In another GENEVA Illumina project (performed by the Broad genotyping center and not described in this article), we evaluated the effect of recalling after removing 846 low quality scans (out of 2970) and replacing them with high quality scans from the same subjects. The discordance between genotype calls in the original and recalled genotypes was extremely low: 1594 of 2124 samples had no discordance and the highest discordance rate was 9 × 10−6.
