Use of Low-Depth GBS Data for Genomic Prediction Across Different Multi-Parental Pools and Single Plants in Perennial Ryegrass (Lolium perenne L.)

A cost efficient genotyping strategy is pivotal for introducing genomic prediction (GP) in plant breeding schemes. Allele-proportions estimates derived by genotyping-by-sequencing (GBS) were efficiently used for GP in periennal ryegrass. Increasing the multiplexing of GBS decreases the genotyping price per unit but also reduces the sequencing depth (SD) per marker and an increased number of missing values. This resuls in a decreased accurancy of the allele-proportion estimation. Both leads to biases in the genomic realtionship matrix (GRM) used in gBLUP GS.

The present work shows methods to estimate and correct the bias affecting the GRM when SNPs at low depth are used. Moreover three methods for imputing missing values, Mean imputation, k-near neighbour, random forest, were tested.

The usefulness of these methodologies to improve the GP accuracies were shown in simulated data with progressively reduced SD. Furthermore, these findings were applied to real data and allowed to efficently combine samples of perennal ryegrass multi-parental genotype-pools and single plants genotype in different assays at different SD. After imputation and bias correction, increased prediction accuracies were observed for single plants (up to +0.05 for heading date), biparental-pools (up to +0.05 for seed yield) and multiparental-pools (up to +0.07 for seed yield).

Our results provide new methodologies to work with GBS estimated allele-proportions which can be obtained for genotype pools or polyploid crops even at moderate or low sequencing depth. Moreover, the feasibility of predicting performances across multi-parental genotype-pools and single plants gave promising perspectives for implementing GP in perennal ryegrass breeding schemes.