Improved Accuracy of Genomic Prediction for Traits with Rare QTL by Fitting Haplotypes

Abstract Text: Genomic prediction estimates QTL effects by exploiting LD. High LD can only occur when SNPs and QTL have similar minor allele frequencies (MAF). Marker panels tend to use SNPs with high MAF and will have limited ability to predict rare QTL. In practice, increasing SNP density has not improved prediction accuracy. This might be explained if a trait had many rare QTL. In such cases, linear models fitting haplotypes could have an advantage because haplotypes could be in complete LD with QTL alleles. SNP genotypes were simulated with 200 SNPs per cM. Genomic breeding values were predicted using either SNP genotypes or non-overlapping haplotypes. When QTL had low MAF, prediction accuracy from haplotype models were significantly higher than for SNP models. Results suggest that haplotype models can be an efficient alternative to SNP models especially when traits are controlled by many rare QTL.

Keywords: prediction accuracy, haplotype, rare variant