Lukas Krusenbaum: Rice gene bank phenomics using publicly available pan-genomic resources

Introduction of novel genetic variation is a key factor in achieving genetic improvements in rice breeding programs. While ample genetic variation from exotic germplasm is stored in gene banks, costs of phenotyping poses a major constraint for utilization of these materials. Genomic prediction (GP) of phenotypes has been proposed to overcome this bottleneck. To ensure sufficient accuracy of GP, appropriate utilization of markers that are representative of the genetic variation causal for phenotypes is needed. We utilized novel Single Nucleotide Polymorphisms (SNPs) identified through a recently published pan-genomic reference database for genomic prediction in gene bank accessions. This database comprises 16 near-gap free reference sequences of Asian rice. By using a published pan-genome graph, we were able to identify SNPs in the dispensable fraction of the rice genome. We found that such SNPs tend to have low call rates (CR) and that their Presence and Absence Variation (PAV) relates to the subpopulation structure of rice. This indicates that a standard approach of handling SNPs by CR filtering and imputation will remove biologically relevant genotypic information from these SNPs. To incorporate this into GP models, we explored modified encodings of SNP variation without CR filtering. These include encoding of PAV and retaining information of nucleotide variation by One-Hot Encoding (OHE). Substantial improvements in prediction accuracy were observed for nine traits in all major subpopulations from incorporating these SNPs under modified encoding. Rethinking pipelines of handling this SNP variation gives an easy-to-apply tool to fully uncover gene bank potential by GP.

Co-Author: Matthias Wissuwa