*3.5. Breeding Application of Genomic Prediction*

Plant breeding is to pyramid favourite alleles from distinct parents using different approaches such as conventional crossing, mutation or transgenic methods to develop new varieties. However, most traits of agronomic importance are genetically controlled by polygenes and have a low heritability such as seed yield and horizontal resistance to diseases. Conventional phenotype selection for these traits is usually inefficient because assessment for them must be performed in multiple environments to obtain breeding values of individuals and thus it is very costly, time consuming and inaccurate; and also because of difficulty of evaluation in fields, greenhouses or laboratories. GS or GP provides an efficient approach to increase selection efficiency by not only increasing selection accuracy but also shortening breeding cycles [58]. In this study, we demonstrate a good example of GP for flax pasmo resistance that is environment-sensitive, costly and difficult for field evaluation. As high as 0.92 of prediction accuracy was obtained for PR, corresponding to 1.84 of relative efficiency over the direct phenotypic selection (Table 3), demonstrating efficiency of GP for low heritability traits. Because the training population underlying the GP models is a diverse germplasm collection that contains more than 90 breeding lines and 245 varieties from different breeding programs [3], the GP models developed in this study are expected to be used for germplasm evaluation, parent selection and individual selection of segregation populations for PR.
