Parallel Genomic Selection with the Island Model: A Simulation Study in Rice

Although genomic selection (GS) is a promising technology, it is necessary to explore efficient ways to incorporate GS in breeding programs. In the present study, we proposed parallel GS with the island model to improve a population derived from multiple-line crosses (MLC), and evaluated its potential with a simulation study. In the simulations, we used real marker data: 3,102 SNP genotypes of 112 Japanese rice (Oryza sativa L.) cultivars. We simulated 100 additive-effect QTL. A GS prediction model was trained using six F₆ populations derived from crosses between six cultivars and one common cultivar. An initial population for GS was consisted of F₂ individuals. Breeding population size was 180. In the parallel GS, a population was separated into six equal-sized subpopulations. GS starting from a population derived from MLC attained higher breeding value than GS starting from a population derived from any of bi-parental crosses. For a population derived from MLC, the parallel GS attained higher breeding value than normal GS. To avoid a strong among-family bias in the first generation of the normal GS, we performed “genome shuffling” via four generations of random mating to “shuffle” genomes derived from different crosses. Normal GS following genome shuffling attained higher breeding value than the parallel GS. The advantage of GS with genome shuffling, however, was not so large as to compensate extra time and effort required for shuffling genomes. The parallel GS with the island model is an efficient way to accumulate beneficial alleles in a population derived from MLC.