Datasets to test the importance of genetic architecture in marker selection decisions for genomic prediction

Abstract

This dataset contains the input files to simulate traits for maize recombinant inbred lines (RILs) and run genomic prediction models with different marker types. Using real genotypic information from 333 maize recombinant inbred lines with single nucleotide polymorphism (SNP) and structural variant (SV) information projected from their seven sequenced parental lines, we simulated traits with different genetic architectures in multiple environments using the R package simplePHENOTYPES. We varied the heritability, the number of quantitative trait loci (QTLs), the type of causative variant (SNPs or SVs), and the variant effect sizes. Weather data from five locations in the U.S. Midwest in 2020 was used to generate a residual correlation matrix among environments. After performing a two-stage analysis with multivariate GBLUP prediction model for each marker type and genetic architecture, we obtained prediction accuracies using two types of cross-validation (CV1 and CV2). For instructions on how to perform this analysis and analysis script, please see https://github.com/HirschLabUMN/genomic_prediction_svs