AI-guided mushroom breeding for agriculture and regenerative manufacturing

Mushrooms can turn agricultural residues into nutritious food and regenerative materials, but breeding better strains is slow and unpredictable. Performance depends on complex genetics interacting with the growing environment, so breeders rely on long cycles of trial and error to find high-yielding, resilient strains suited to local conditions.

This project is building a faster, more predictive path, grounded in virtual mushrooms. We build foundation models that learn from DNA, protein, and phenotypic data, such as imaging, to connect a mushroom’s genome to its traits, and from cultivation and production data to optimize strain growth. We learn quickly in a fast-growing model mushroom, then test how much this knowledge transfers to crops like oyster mushrooms. We work with a co-mycologist, a reasoning agent that assists breeders by proposing crosses and growing conditions most likely to achieve a goal and explaining why, with the underlying predictions coming from the models. The aspiration is inverse design: to name a target, such as higher yield, resilience, shelf life, or flavor, and have the models propose the crosses most likely to achieve it.

The goal is to reduce the time and cost of developing improved strains, helping to turn agricultural residues into reliable, nutritious food and sustainable materials. We use collaborative training techniques to improve model performance while protecting breeders’ proprietary information. Because these models learn how fungi grow and develop, the same foundation can extend beyond mushroom breeding to fungal biotechnology more broadly.