Evaluating genomic prediction accuracy for tuber yield and quality traits in tetraploid potato across single and multiple year predictions (P-04)

The University of Maine Potato Breeding Program develops improved varieties to benefit growers in Maine, the U.S., and beyond. Traditional breeding relies on phenotypic recurrent selection, which is effective but slow - up to 12 years to release a cultivar. Potato polyploidy and high heterozygosity further complicate breeding value prediction. Genomic selection (GS) uses genome-wide markers to estimate genotypic merit and has the potential to accelerate breeding and enhance selection accuracy.

This study evaluated the prediction accuracy of a GS model using the GBLUP approach, considering both within-year and across-year schemes relevant to breeding applications. The third-year population from the UMaine program was phenotyped for total and marketable yield, specific gravity, and chip color from 2020 to 2024, totaling 971 clones. All clones were genotyped with the Illumina SNP array v4 (~31k markers); after quality filtering, 15k markers were retained. The G-matrix was estimated using the AGHmatrix R package, and models were fitted in ASReml-R v4.0.

In the first scenario, cross-validation (CV) was performed independently within each year, with 90% of clones used for training and 10% for validation. A second CV scheme used one year for training to predict the next, evaluating the feasibility of across-year predictions. Training population size was further increased by combining data from multiple years to predict the following year. These approaches assess the model’s ability to predict untested genotypes—a key advantage of GS in reducing breeding cycles and supporting more efficient selection.

The narrow-sense heritability estimates range from 0.25 to 0.58 for marketable yield and 0.38 to 0.85 for specific gravity. Our preliminary results show prediction accuracy from 0.13 to 0.59 and 0.32 to 0.58 for marketable yield and specific gravity, respectively. We will discuss the results of additional CV scenarios and discuss the implications of these findings for enhancing genomic selection strategies.