CVRMS: Cross-validated Rank-based Marker Selection for Genome-wide Prediction of Low Heritability

Abstract

CVRMS is an R package designed to extract marker subsets from repeated rank-based marker datasets generated from genome-wide association studies or marker effects for genome-wide prediction (https://github.com/lovemun/CVRMS). CVRMS provides an optimized genome-wide biomarker set with the best predictability of phenotype by implemented ridge regression using genetic information. Applying our method to human, animal, and plant datasets with wide heritability (zero to one), we selected hundreds to thousands of biomarkers for precise prediction.Genomic prediction (GP) (Meuwissen et al., 2001) is a potent tool in plant and animal breeding and in the prediction of complex traits and disease risks in humans. Several GP methods, such as the method of marker effect (Meuwissen., 2001;Habier et al., 2011;Gianola, 2013) and genomic best linear unbiased prediction (GBLUP) (VanRaden, 2008), have been proposed over the years.A traditional statistical prediction approach is that large numbers of predictor variables, such as SNPs, are regressed on phenotypes. Recently, as SNP density and sample size have increased and