Evaluation of SNP Genotyping Errors Using Quality-Assurance Sample Checks with Implications on Genomic Prediction

Genotyping errors are of serious concern in applications such as map construction, genome-wide association studies, and genomic selection. A genotype error occurs when an observed genotype does not correspond to the true underlying genotype. Population-wise, genotype errors which are inconsistent with Mendelian inheritance can be detected and eliminated, yet there may still exist many Mendelian-consistent genotype errors which are hard to be detected. From the perspective of quality assurance (QA), genotype errors can be identified by repeatedly genotyping a pre-selected set of animals (i.e., QA sample checks) on the same or different chips.

In the present study, genotype error rates were estimated for some commonly-used commercial SNP chips (GGPLDV3, GGPLDV4, GGPHD80K, GGPHD150K, and GGPHD777K) based on replicated genotypes for QA sample checks in GeneSeek (a Neogen company). True genotypes for these QA sample checks were called by majority voting with the corresponding probabilities inferred. On average, genotype error rates were low (0.02% - 0.08%) for these SNP chips. Chromosome-wide, genotype error rates were mostly less than 0.05%, but significantly elevated genotype errors were observed with SNPs on Y chromosome (0.028 – 4.786%) and those on the Mitochondrial DNA (0.179 – 0.375%). Observed genotype error rates varied with samples, chromosomes and chips. The differences might be due to varied DNA quality, categories of SNPs, and genotyping platforms (including chemicals). The impact of genotype errors on genomic prediction was investigated in a Holstein population by randomly simulating average numbers of genotype errors for SNPs on the GGPHD (80K) chip. The results showed only slightly decreased genomic prediction accuracy. Hence, the present SNP genotype error rates are not of great concern in applications such as genomic prediction, provided that the genotype error rate are kept at low levels. The robustness of genomic prediction accuracy depends on the whole system rather than heavily on individual SNPs. Finally, a list of SNPs with a genotype error rate great 1% was provided. These SNPs should be interpreted with caution in applications (such as genome-wide association studies) in which high accuracies of single SNPs are expected.

* Correspondence: nwu@neogen.com