The data-driven approach of startup Heritable Agriculture aims to modernize traditional plant breeding methods. Recently spun off from Google's X-Division, the company combines machine learning with plant genomics to predict breeding outcomes more accurately.
The company's AI platform goes beyond simple genetic analysis. It examines how genes interact with other plant molecules (multi-omic data), making it easier to pinpoint which genes control specific traits. Their models work similarly to language models, except they process DNA sequences instead of words to identify genome sections controlling plant characteristics.
The system also connects genetic information with environmental factors like weather, soil quality, and climate to predict how well specific plant variants will grow in different locations. This approach could reduce the need for lengthy field trials, potentially accelerating the development of new plant varieties.
Data collection on an industrial scale
At the heart of their work was a special test chamber at X headquarters, where an automated camera system recorded the plant's development every hour. The system recorded precise measurements such as flowering times and structural changes, helping the team to validate their AI models.
To build their database further, the team conducted field trials across California, Wisconsin, and Nebraska. They collected detailed measurements, from counting corn kernels per cob to measuring vegetable bitterness levels, preserving samples in liquid nitrogen for analysis.
"We’re not developing gene-edited plants, and genetic modification is not on our roadmap," CEO Brad Zamft tells TechCrunch. For now, the startup focuses on using AI analysis to identify optimal crosses for conventional breeding, though Zamft suggests gene editing might be considered later.
