Cracking the code
How AI is helping predict hybrid corn performance before planting
Predicting how a new corn hybrid will perform in the field is one of agriculture’s toughest puzzles – now a breakthrough AI model is helping solve it.
As an example, take a hybrid created in Brazil: will it thrive in another tropical climate like Thailand? For decades, the only reliable way to answer that question was to plant it and wait.
But not anymore. Syngenta’s Seeds Digital R&D team has designed and deployed an advanced AI model known as “Pyrgos” that combines genetics, environmental conditions, and agronomic practices to forecast crop performance with unprecedented accuracy.
These different factors also make predicting the yield of a particular seed a daunting challenge but with Pyrgos, scientists can better predict product placement and performance in advance, giving more reliable agronomic management practices before a single seed goes into the ground.
Eric Ginsberg is the lead scientist on the project: “As a company we have this incredible resource of trial data from around the world, so we started by asking: how could we bring forward new statistical techniques and new technology to unlock new insights?”
In essence, the model can simulate how any corn hybrid would perform in any environment – from a rainy valley somewhere in Europe to a dry prairie in the American mid-west.
To do this, Pyrgos considers hundreds of different environmental factors (like weather patterns and soil types) alongside each hybrid’s unique genetic profile.
By learning from millions of data points across more than one million distinct genetic hybrids, the model detects subtle patterns and interactions that traditional tools could never capture.
“Our vision was to set a new standard in large-scale G×E modeling,” explains Cole Manring, a machine learning engineer. “We applied advanced AI to ask some big ‘what-if’ questions of our entire trial database – and in return, we’re getting insights that drive big decisions.”
“For the first time, we have an AI model that can tell us how a hybrid from one region might perform in another,” says Clay Cole, an R&D Digital Advancement & Placement Lead at Syngenta. “These insights help us share high-performing germplasm across regions more efficiently, increasing genetic diversity in our pipeline and get elite products to farmers around the world more quickly.”
The name “Pyrgos” comes from the Greek word for “tower” – a nod to the model’s two-pillar design. It is this architecture that allows Pyrgos to draw on data from one region (say, Brazil) to predict outcomes in another (like India) with appropriate calibrations.
Initial results are promising: Pyrgos outperforms previous prediction methods and is especially useful when forecasting the performance of completely new hybrids that have never been grown in a particular environment.
In practical terms, that means when plant breeders create a brand-new corn hybrid, Pyrgos can accurately estimate its future yield and traits – even under conditions the hybrid hasn’t experienced before.
Andy Jakubowski, Head of Seeds R&D Digital at Syngenta, is excited about the possibilities that Pyrgos opens up.
Building Pyrgos was a truly global effort. Eric says: “This required working in a new way and joining up data to make it useful.” Data scientists, agronomists, plant breeders, and software and data engineers collaborated to ensure the model not only pushes the envelope of AI but also delivers actionable results on the ground.
By making a major leap forward towards cracking the code on GxExM interactions Syngenta can develop crop products that are more precisely matched to local growing conditions, leading to better and more reliable performance in the field.
