Corteva Accelerates Seed and Crop Protection Discovery With NVIDIA AI

The majority of plant diseases are caused by pathogenic fungi. Corteva is focused on understanding how these pathogens interact with plant proteins, mapping host–pathogen interactions at scale to better understand the biological mechanisms that control disease.

“Whenever a pathogen is attacking the plant, it deploys proteins, and it’s a battle,” said Abhiman Saraswathi, Data Science Manager at Corteva. “At the same time, the plant is deploying its defense mechanisms to kill the pest and ward off the infection—these are the protein interactions we want to map and understand.” 

The challenge: there are approximately a billion possible protein interactions between pathogens and host plant proteins. 

Corteva collaborated with NVIDIA to implement BioNeMo Boltz-2 NIM, a high-performance inference microservice for biomolecular structure and binding affinity predictions, to help create this massive interaction map. The team is also using BioNeMo Boltz-2 NIM to verify the occurrence of these interactions. 

Specifically, this work focuses on understanding interactions that drive the most impactful crop diseases, which can cause annual crop yield losses of 10–20%.

To identify these interactions, the team goes through the following workflow pipeline: 

1. Identify which pathogen proteins have features in their sequence, which means they are participating in pathogenesis and an interaction is likely occurring with a host protein.

2. Use BioNeMo Boltz-2 NIM to curate the interaction map for this specific subset, which, according to Corteva’s internal benchmarks, is better at constructing these complex interactions between pathogens and host proteins than other co-folding models.

3. Develop workflows that can handle the computational scale of billions of interactions using accelerated computing.

After benchmarking the BioNeMo Boltz-2 NIM model for speed and scale using NVIDIA H200 GPUs, the model has been proven to be 4.15x faster than the open-source model. This increase in speed and scaling enables mapping the protein interaction network in weeks instead of years, while maintaining the accuracy of the results.

This work is focused on early-stage biological discovery, generating insight that can inform multiple downstream approaches to improving plant resilience.

Globally, 20–40% of crop production is lost each year due to weeds, pests, and disease, creating ongoing pressure on farmers to protect yield while meeting sustainability expectations. Corteva is applying AI and accelerated computing to its natural products platform to more efficiently identify and prioritize high-value molecules from increasingly large and complex datasets. This capability directly supports Corteva’s crop protection pipeline by accelerating how natural product candidates are discovered and evaluated.

Natural products are a cornerstone of Corteva’s crop health R&D, providing a powerful foundation for developing sustainable solutions that perform in the field and meet modern environmental expectations. 

Corteva has a massive, continuously growing mass‑spectrometry dataset to profile microbial natural products. “The real key problem is that this is a needle in a haystack situation at a massive scale,” said Deepa Acharya, Discovery Metabolomics Leader at Corteva. “Microbes make thousands of metabolites, and our mission is to find the one metabolite that can truly protect the crops and become a product.”

The Corteva research team—in partnership with the Wang Bioinformatics Lab at the University of California, Riverside—needed to analyze the huge datasets fast enough to extract actionable biological insights. The key bottlenecks were the clustering of mass spectrometry data and generating large molecular networks. 

This collaboration reimagined this computational metabolomics workflow. By combining algorithmic insights from the Wang Lab, including incremental clustering—with NVIDIA toolkits such as RAPIDS cuML for clustering and RAPIDS cuVS for networking—the team experienced significant acceleration of the workflow.

The end-to-end metabolomics workflow resulted in an 80-fold to 100-fold gain in speed in mass spectral clustering, 50-fold to 60-fold in embedding, and 15-fold in molecular networking. Enabled by the acceleration, the Corteva team can now run the workflow more frequently, from quarterly to monthly, leading to more natural product discoveries.

Repository molecular network, where each node represents a metabolite and edges encode similarity. Colored nodes denote annotated compounds classified by type, while black nodes represent unannotated compounds yet to be identified.

This research is the building block for developing a deeper understanding of organisms to further unlock natural solutions. 

“The more we can profile complex organisms and samples, and cluster that data together to find patterns—the more impactful metabolomics analysis becomes across industries, including agriculture technology, biotechnology, pharmaceutical, and beyond,” said Acharya.

The impact of these capabilities is strengthened by Corteva’s proprietary datasets, domain expertise, and integration into its R&D pipeline, making them difficult to replicate through compute alone.

Generative AI is unlocking the crop protection and seeds universe for agriculture. Learn more about how NVIDIA CUDA-X libraries for data science and BioNeMo can accelerate your research and discovery pipelines.