Why AI is the New Frontier in Herbicide Development
In a bold move to tackle the twin challenges of herbicide resistance and human health risks, the U.S. Advanced Research Projects Agency for Health (ARPA-H) has awarded up to $53 million to a consortium led by the Massachusetts Institute of Technology. The funding will fuel a multi-institution effort to build generative-AI tools that can design next-generation weed killers—molecules or RNA-based agents—that are both more effective against resistant weeds and safer for people and the environment.
The initiative, announced by MIT News, marks a rare convergence of biotechnology, chemical engineering, and cutting-edge machine learning, all under the umbrella of ARPA-H’s STREAM (Systems for Tracking and Resilience in Efficient Agricultural-input Management) program. By leveraging AI to explore billions of chemical possibilities, the team hopes to reduce U.S. reliance on imported phosphorus, the key ingredient in glyphosate, while delivering herbicides that mitigate the health concerns linked to current agrochemicals.
Direct answer
ARPA-H is funding a $53 million MIT-led project that will use generative artificial intelligence to design new herbicide molecules and RNA-based agents, aiming to overcome weed resistance, lower human health risks, and cut dependence on imported phosphorus.
The Scientific Imperative Behind the Funding
More than 1 billion pounds of herbicide are applied annually in the United States. These chemicals protect crops from weed competition, enabling high yields and supporting regenerative farming practices that preserve soil health. However, the widespread use of herbicides such as glyphosate and atrazine has created two pressing problems:
- Resistance mutations: Weeds are evolving genetic changes that render existing herbicides ineffective, threatening crop productivity.
- Human health concerns: Exposure to herbicides has been linked to increased risks of cancer, metabolic disorders, and endocrine disruption, affecting farm workers and the broader public through food and water.
These issues intersect with national-security worries. The majority of phosphorus, a critical component of glyphosate, is imported, making the supply chain vulnerable. As Christopher Voigt, head of MIT’s Department of Biological Engineering, notes, “If you combine the resistance mutations with issues around national security of the supply chain of the precursors to herbicides, we could find ourselves without these core agrochemicals that are absolutely essential for getting the high crop yields and economical prices that the United States now sees.”
Project Structure: The PLOW Consortium
The effort, officially titled Precision Learning for On-target Weed control (PLOW), assembles expertise from six institutions:
- MIT (lead institution)
- Broad Institute of MIT and Harvard
- University of North Carolina at Chapel Hill
- University of Tennessee at Knoxville
- University of Illinois at Urbana-Champaign
- Robigo (a biotech startup)
Researchers are organized into three technical teams, each aligned with a STREAM technical area.
Team 1: Generative-AI Model Development
Led by Connor Coley (MIT Chemical Engineering), Regina Barzilay (MIT AI and Health), and Alexander Tropsha (UNC Eshelman School of Pharmacy), this team will create machine-learning models capable of predicting candidate molecules or small interfering RNAs (siRNAs) that target the same plant enzymes as existing herbicides, most notably EPSP synthase—the enzyme glyphosate blocks to halt amino-acid synthesis.
The models will be trained to optimize multiple objectives simultaneously:
- High potency against target enzymes
- Low likelihood of resistance development
- Minimal toxicity to humans and non-target organisms
- Cost-effective synthesis pathways
Because generative AI can explore a combinatorial space far beyond human intuition, the team expects to generate “billions of possible candidates,” which will then be filtered through experimental testing.
Team 2: High-Throughput Screening
Angela Koehler (MIT Biological Engineering) and Jaime Cheah (Broad Institute) will head the screening effort. Their laboratory will synthesize or assemble the AI-suggested candidates and evaluate them for target engagement, toxicity, and resistance risk. Crucially, the screening data will be fed back into the AI models, creating a closed-loop learning system that continuously refines predictions.
The first round of greenhouse tests for RNAi-based candidates is slated for completion by the end of year one, providing early validation of the AI-driven pipeline.
Team 3: Greenhouse Validation and Microbial Degradation
The final validation stage will be conducted by a coalition of plant scientists and synthetic-biology experts, including Neal Stewart and Scott Lenaghan (University of Tennessee), Patrick Tranel (University of Illinois), Tim Willson (UNC Pharmacy), and Andee Wallace (Robigo). Their work will assess how promising molecules perform under realistic agricultural conditions, measuring efficacy, crop safety, and environmental impact.
In parallel, the consortium will engineer microbial systems—bacteria, fungi, and synthetic RNA—that can degrade residual agrochemicals before they enter food chains or waterways, offering a bioremediation layer that could further reduce human exposure.
AI Techniques at Play
While the source does not detail specific algorithms, the involvement of Barzilay and Coley suggests the use of state-of-the-art generative models such as variational autoencoders, graph-based neural networks, and reinforcement learning frameworks that have already shown promise in drug discovery. These methods excel at navigating high-dimensional chemical space, proposing structures that satisfy multiple constraints.
What distinguishes the PLOW effort from typical AI-driven chemistry projects is the explicit inclusion of resistance-mitigation criteria. By modeling potential resistance pathways—e.g., mutations in EPSP synthase—the AI can prioritize molecules that retain activity even when weeds evolve.
Potential Impact on the Agricultural Landscape
If successful, the project could reshape several aspects of U.S. agriculture:
- Supply-chain resilience: Reducing dependence on imported phosphorus would insulate farmers from geopolitical disruptions.
- Health and safety: Safer herbicide chemistries or RNA-based agents could lower occupational exposure for farm workers and reduce trace residues in the food supply.
- Resistance management: Molecules designed to be less prone to resistance could extend the useful lifespan of chemical weed control, delaying the need for alternative tactics.
- Environmental stewardship: Engineered microbes that degrade agrochemicals could mitigate runoff, protecting waterways and biodiversity.
These outcomes align with broader national goals of food security, sustainable farming, and climate-smart agriculture. Moreover, the project exemplifies how federal research agencies are increasingly betting on AI to accelerate the discovery of solutions to entrenched biological challenges.
Challenges and Uncertainties
Despite the promise, several hurdles remain:
- Regulatory pathways: New herbicide classes—especially RNA-based agents—will need to navigate rigorous EPA and FDA assessments, which can be time-consuming.
- Scale-up feasibility: Even if AI identifies a low-toxicity molecule, manufacturing at agricultural scale must be cost-effective.
- Ecological complexity: Predicting long-term ecosystem impacts of novel chemicals or engineered microbes is inherently uncertain.
- Data integration: The feedback loop between high-throughput screening and AI models relies on high-quality, standardized datasets, which can be difficult to curate.
These factors mean that, as of now, no performance data or field results have been published; the project is still in the early research phase.
Broader Context: AI in AgTech
The PLOW consortium joins a growing cadre of initiatives that apply generative AI to agricultural inputs. Similar efforts in pesticide design, fertilizer optimization, and crop-trait prediction are leveraging the same computational paradigms. What sets PLOW apart is its explicit focus on human health and national security, reflecting a policy-driven urgency that goes beyond pure yield maximization.
In the Fourth Industrial Revolution, the integration of AI with synthetic biology is reshaping how we approach problems that were once solved solely through trial-and-error chemistry. The success—or even the lessons learned—from this $53 million investment will likely inform future public-private partnerships aimed at securing the food supply while safeguarding the environment.
FAQ
What is the total amount of funding provided by ARPA-H for this project?
Up to $53 million has been allocated to the MIT-led consortium.
Which agencies and institutions are involved in the research?
The effort is led by MIT and includes the Broad Institute of MIT and Harvard, the University of North Carolina at Chapel Hill, the University of Tennessee at Knoxville, the University of Illinois at Urbana-Champaign, and the biotech company Robigo.
What are the main goals of the PLOW project?
To develop generative-AI tools that design herbicide molecules or RNA agents which are more effective against resistant weeds, have lower human toxicity, reduce reliance on imported phosphorus, and can be degraded by engineered microbes.
How will AI be used to create new herbicides?
AI models will predict billions of candidate compounds or siRNAs that target key plant enzymes, optimizing for potency, low resistance risk, low synthesis cost, and safety. Screening data will be fed back to improve the models iteratively.
When are the first greenhouse tests expected?
Initial RNAi candidates are scheduled to complete greenhouse testing by the end of the first year of the program.
Will the new herbicides be chemical or biological?
The project explores both chemical molecules and small interfering RNA (siRNA) agents, as well as microbial systems that can degrade agrochemicals.
What health concerns are associated with current herbicides?
Exposure to existing herbicides has been linked to increased risks of cancer, metabolic and endocrine disorders, and other adverse health outcomes for farm workers and the general public.
Conclusion
The ARPA-H-funded PLOW initiative represents a decisive step toward marrying artificial intelligence with agricultural chemistry. By confronting herbicide resistance, health risks, and supply-chain fragility in a single research agenda, the consortium could deliver a new class of crop-protecting agents that are both scientifically sophisticated and socially responsible. While the work is still in its infancy, the collaborative, data-driven approach exemplifies how the Fourth Industrial Revolution can be harnessed to secure a more resilient and healthier food system for the future.
Source: MIT News
