Impact of genome assemblies, genotyping methods, variant types, ploidy levels and population structures on genomic prediction in switchgrass

Genomic prediction, where genotype information is used to predict phenotypes, has accelerated the breeding processes and can provide mechanistic insights into phenotypes of interest. Switchgrass (Panicum virgatum L.) is a perennial biofuel feedstock with multiple traits targeted for accelerated breeding using genomic prediction approaches. To optimize switchgrass genomic prediction, we assessed the impact of genome assembly versions, sequencing strategies for variant calling, variant types, variant allelic complexities, and polyploidy levels on the prediction of 20 traits in a switchgrass diversity panel with 486 individuals. We found that genome assembly has limited impact on prediction accuracy. Bi-allelic insertion/deletions and multi-allelic variants are as useful as bi-allelic single nucleotide polymorphisms. In addition, models built using exome capture-derived variants tend to have higher prediction accuracy than those using genotyping-by-sequencing variants. Sequencing depth, ploidy levels and population structures also have significant effects on prediction accuracy. The prediction accuracy can be improved by integrating different types of variants. We also show that the anthesis date prediction models based on exome capture variants, especially those using exome capture multi-allelic indels, identified the highest numbers of genes similar to known flowering time genes in other species. Our study provides insights into the factors influencing genomic prediction outcomes that inform best practices for future studies and for improving agronomic traits in switchgrass and other species through selective breeding.