Genotyping errors, pedigree errors, and missing data

Our group studied the effects of genotyping errors, pedigree errors, and missing data on a wide range of techniques, with a focus on the role of single-nucleotide polymorphisms (SNPs). Half of our group used simulated data, and half of our group used data from the Collaborative Study on the Genetics of Alcoholism (COGA). The simulated data had no missing genotypes and no genotyping errors, so our group, as a whole, removed data and introduced artificial errors to study the robustness of various techniques. Our teams showed that genotyping errors are less detectable and may have a greater impact on SNPs than on microsatellites, but recently developed methods that account for genotyping errors help reduce false positives, and the assumptions of these methods appear to be supported by observations from repeated genotyping. The ability to detect linkage disequilibrium (LD) was also substantially reduced by missing data; this in turn could affect tagging SNPs chosen to generate haplotypes. In the COGA sample, genotyping measurements were repeated in three ways. First, full-genome screens were performed on three sets of markers: 328 microsatellites, 11,560 SNPs from the Affymetrix GeneChip Mapping 10 K Array marker set, and 4,720 SNPs from the Illumina Linkage III panel. Second, the entire Affymetrix marker set was typed on the same 184 individuals by two different laboratories. Finally, the Affymetrix and Illumina marker panels had 94 SNPs in common. Our teams showed that both SNPs and microsatellites can be readily used to identify pedigree errors, and that SNPs have fewer genotyping errors and a low inconsistency rate. However, a fairly high rate of no-calls, especially for the Affymetrix platform, suggests that the inconsistency rate may be higher than observed.