Genomic prediction models based on a large-scale recombinant population allow quick breeding of high-yield rice
Rapid development of cultivars with optimal genome-wide allele combinations is essential for addressing agricultural challenges. However, constructing desired genotypes through conventional cross-breeding requires many generations, calling for a more efficient methodology. Here, we present a rapid breeding strategy that combines a large-scale recombinant population as starting material with interpretable genomic prediction models trained on that population. This approach enables efficient construction of target genotypes optimized for multiple traits in cultivars. To validate our strategy in rice (Oryza sativa), we established a nested association mapping population of 2,787 recombinant inbred lines derived from an elite cultivar 'Hitomebore' and 19 diverse donors. We built highly accurate genomic prediction models using this population. We then used the models to estimate haplotype-specific effects and account for trade-offs among yield-related traits, and selected optimal lines and designed breeding schemes. Crossbreeding based on these schemes produced rice lines with the target genotypes for multiple yield-related traits, supporting the predicted effects and validating the effectiveness of our strategy. This genomic breeding approach provides a general framework for rapidly breeding cultivars able to meet the challenges posed by a changing environment.