Genomic prediction models based on a large-scale recombinant population allow rapid breeding of desired genotypes

For centuries, farmers have relied on a slow, patient art to improve their crops: cross-breeding. By mating two plants with different strengths, they hope to combine those traits in the offspring. But nature is unpredictable, and the genetic lottery of traditional breeding often requires many generations of planting, waiting, and selecting before a truly superior variety emerges. In an era where climate change and population growth demand faster solutions, this slow pace is a bottleneck. Scientists have long sought a way to look inside the plant's genetic code to predict which combinations will work best before a single seed is even planted. This approach, known as genomic prediction, uses computer models to estimate how specific pieces of DNA will influence a plant's growth, yield, or resilience. The challenge has been finding a way to make these predictions accurate enough to guide real-world farming, especially when trying to balance multiple goals at once, such as making a crop both high-yielding and drought-resistant.

A team of researchers has now demonstrated a powerful new strategy that bridges the gap between complex genetic data and practical farming. They began by creating a massive library of genetic variety, a collection of 2,787 distinct plant lines derived from crossing a top-performing rice variety known as 'Hitomebore' with 19 other diverse rice donors. This large group, called a nested association mapping population, served as a detailed map of how different genetic pieces interact. Using this extensive dataset, the scientists built highly accurate computer models capable of reading the genetic code of these lines and predicting their performance. Unlike previous methods that might guess at the outcome, these models were trained on the specific genetic combinations present in the study, allowing them to estimate the effect of specific DNA segments with remarkable precision.

The true test of this strategy came when the researchers used their models to design a breeding plan from scratch. Instead of relying on chance, they asked the computer to identify the specific genetic combinations that would maximize yield while managing the natural trade-offs between different traits. The models pointed them toward a set of optimal lines and a specific path to cross them. Following this digital blueprint, the team performed the actual breeding work, crossing the selected plants to create new generations of rice. The results confirmed the power of the approach: the new rice lines possessed the exact target genotypes the models had predicted, delivering the desired improvements in multiple yield-related traits simultaneously. This work shows that by combining a large, diverse starting population with interpretable prediction models, breeders can rapidly construct cultivars with the precise genetic makeup needed to meet the challenges of a changing world.