A training population to evaluate the accuracy of genomic selection for fry colour and resistance to low temperature sweetening was established over the two years of this action. This was made up of potato entries at the year five stage of evaluation in the breeding programme, where the number of candidate varieties has gone from just under 100,000 lines to approximately 300 lines. The final training population consisted of 450 entries that were evaluated for fry colour ‘off-the-field’, and at various time points during storage at 4.50C and 80C (with sprout suppressant) . In total 9,440 tubers were sliced into crisps, deep fried, and analysed for fry colour using a HunterLab LabScan XE spectrophotometer. Results of this enhanced phenotyping effort have already been exploited for parental selection and advancement of lines within the breeding programme. The training population was also genotyped using a genotyping-by-sequencing approach and a database of 129,008 Single Nucleotide Polymorphisms (SNPs) was established, which characterised genetic variation across the genome within the training population. These data-sets were brought together and used to evaluate the accuracy of genomic selection for fry colour and resistance to low temperature sweetening. The aim was to build predictive models with genome-wide SNP data and determine how well these data can predict our traits. We evaluated various statistical algorithms and determined factors affecting predictive ability (e.g. training population size, marker density, and relationship between training and testing sets). Predictive ability was high (ranging from 0.61 to 0.72) when predicting fry colour at various time points during storage and reducing SNP number had limited impact on predictive ability. Much of the predictive ability was due to SNPs capturing familial relationships, and predictive ability dropped when material unrelated to the training set was included in the testing set. However, we also performed a genome-wide association study to identify individual SNPs associated with fry colour. This differs from genomic selection in that we are testing each marker in turn for association with the trait after correcting for structure in the training population. This enabled the identification of a subset of molecular markers that together were capable of predicting fry colour and resistance to low temperature sweetening with high accuracy. Identified markers had the same predictive ability as the entire marker set and greater predictive ability than a similar number of randomly selected markers, indicating they are in linkage disequilibrium with quantitative trait loci. Work initiated in this action will continue at the hosting organisation for the foreseeable future. All findings and data from this action will be made available via open access publications and according to FAIR principles.
