Correction for Measurement Error from Genotyping-by-Sequencing in Genomic Variance and Genomic Prediction Models

Genotyping-by-sequencing (GBSeq) is becoming a cost-effective genotyping platform for species without available SNP arrays. GBSeq considers to sequence short reads from restriction sites covering a limited part of the genome (e.g., 5-10%) with low sequencing depth per individual (e.g., 5-10X per sample).  The GBSeq data can be used directly in genomic models in the form of individual SNP allele-frequency estimates (e.g., reference reads/total reads per polymorphic site per individual), but is subject to measurement error due to the low sequencing depth per individual. Due to technical reasons, sequencing depth also varies between SNPs, so that some SNPs suffer much more from measurement error than others. Elsewhere, in the context of association studies, we have shown that this measurement error leads to underestimation of allele effects and we derived a correction for this underestimation. In the current work we show how the correction for measurement error in GBSeq can also be applied in whole genome genomic variance and genomic prediction models. Bayesian whole-genome random regression models are proposed to allow implementation of large-scale SNP-based models with a per-SNP correction for measurement error. We show correct retrieval of genomic explained variance, and improved genomic prediction when accounting for the measurement error in GBSeq data.