Genotyping by sequencing (GBS) allows generating up to millions of molecular markers with a cost per sample which is proportional to the level of multiplexing.

Increasing the sample multiplexing decreases the genotyping price but also reduces the numbers of reads per marker. In this work we investigated how this reduction of the coverage depth affects the genomic relationship matrices used to estimated breeding value of F2 family pools in perennial ryegrass.

A total of 995 families were genotyped via GBS providing more than 1.8M allele frequency estimates for each family with an average coverage depth of 12.6 per marker. Simulated datasets with a progressively reduced depth showed an increasing level of missing values together with an overestimated genetic variance caused by inflated diagonals in the genomic relationship matrix.

In order to address these drawbacks we first showed how to correct the diagonal elements by estimating the amount of genetic variance caused by the reduction of the coverage depth. Secondly we developed a method to scale the relationship matrix by taking into account the overall amount of pairwise non-missing loci between all families.

Rust resistance and heading date were chosen as example traits to prove that these two procedures can considerably mitigate the loss of accuracy in predicted breeding values associated with the decreasing coverage depth. These findings will allow increasing the sample multiplexing in the GBS assays reducing the genotyping cost or increasing the size of the population under analysis for the same cost.