The agricultural industry is struggling with increasing demand for more yield with better resource management and the looming pressure of climate change. The International Rice Research Institute (IRRI) produces rice hybrids and varieties for many different countries and climates globally. IRRI has yield measurements for 3,000 rice varieties but only from the Philippines. It is impractical for IRRI to test the performance of all rice varieties in all of these different locations.

The environment and the interplay of genes have a major effect on many important traits such as yield or flowering time. With changing climate, the question to answer therefore is how these rice varieties perform in different weather conditions and locations – a question that not only concerns rice breeders, but also producers of almost all crops.

IRRI has chosen Computomics’ machine learning-based technology xSeedScore to speed up data analysis and advance rice breeding efforts. Our goal is to optimize the development of improved rice varieties to deliver increased yields, better nutritional content, or stronger resilience to environmental stresses brought about by climate change in untested locations and climates.

xSeedScore’s machine learning technology models non-linear interactions between genotype and environment (GxE), between genes (GxG), and can also factor in farming practices (GxExM). By taking into account all these phenotype influencing factors, xSeedScore builds a model that is a much better approximation of the biological reality than what traditional technologies offer.

xSeedScore integrated extensive weather data – such as temperature, precipitation, wind speed, soil characteristics, humidity and solar – for multiple years, in 15 different locations and during different seasons. An all-encompassing machine learning model predicted the phenotypes of over 6.600 plants for yield, flowering time and height, and surpassed prediction accuracies of traditional methods. By modelling the environment-genotype interaction, xSeedScore could not only predict the best plant for a certain environment. By forecasting future climate, we can support breeders in deciphering the best genetic makeup of plants for a certain region 10 years in the future.

Our solution achieves higher prediction accuracies by incorporating environmental measurements and predicts plant performance in untested locations and future climates. xSeedScore gives recommendations as to which crop performs well at a certain location and is therefore of high value for crop placement. Selecting the right variety favorable to changing climate conditions is critical to food security.

Improved crop varieties benefit not only breeders, but ultimately farmers and consumers around the world.