We are building language-model approaches that help connect DNA edits to gene function and plant performance. By learning the causal impact of genetic changes, these models can guide the design of crops with greater climate resiliency, improved efficiency, and sustainable production pathways for the future bioeconomy.
Modeling from DNA sequence to performance across environments is a high-dimensional, nonlinear learning problem. Off-the-shelf AI alone is not enough. We build gray-box models that combine domain knowledge with data-driven learning to improve sample efficiency and deliver interpretable insights.
Agriculture advances in yearly cycles, yet genetic gain must accelerate to meet growing global demand. We translate ideas from Bayesian experimental design and optimization into the agriculture context to prioritize the most informative experiments and speed discovery.
Mean-squared error is rarely the best objective for agricultural design. We develop loss functions that focus learning on what matters most: rare discoveries, cell-type specificity, and targeted populations or alleles. The goal is to learn what we need for actionable design, not everything possible.
Agricultural design spans billions of base pairs, multiple phenotypes, complex environments, and economic constraints. No single model solves it all. We build agentic workflows that explore more options, navigate trade-offs, and move decisions forward faster while keeping insights transparent and actionable.