An MIT-led research team will use funding of up to $53 million to pursue new ways of controlling weeds. The effort will combine generative artificial intelligence, laboratory screening, and greenhouse studies to look for alternatives designed with crop protection and human health in mind.
The work responds to several pressures described by the researchers: weeds are developing resistance to common herbicides, exposure may carry health risks, and reliance on imported phosphorus used in glyphosate raises supply concerns.
Why researchers are looking beyond current herbicides
Herbicides help farmers prevent substantial crop losses from weeds and support regenerative farming practices that protect soil. But repeated use has allowed some weeds to develop genetic mutations that make commonly used products less effective, threatening the role these chemicals play in crop production.
Glyphosate works by interfering with EPSP synthase, an enzyme plants need to produce several critical amino acids. The project will investigate molecules that disrupt the same plant enzymes targeted by glyphosate and other herbicides, while seeking candidates that could help prevent resistance from emerging.
Health and environmental concerns are another part of the motivation. The source article says herbicide exposure may contribute to increased risk of cancer, metabolic and endocrine disorders, and other adverse health outcomes. Farm workers may encounter chemicals through frequent use, while many U.S. residents are exposed to small amounts through food and drinking water.
AI will propose candidates for testing
The project, called Precision Learning for On-target Weed control (PLOW), is funded by the U.S. Advanced Research Projects Agency for Health (ARPA-H) through its STREAM (Systems for Tracking and Resilience in Efficient Agricultural-input Management) program. Christopher Voigt, head of MIT’s Department of Biological Engineering, is the principal investigator.
One team will develop generative AI models to predict possible herbicide candidates. These could be chemical compounds or small interfering RNA (siRNA). The models are intended to optimize for effectiveness as well as low synthesis cost and low risk of resistance, harm to humans, or ecological damage.
The AI group is led by Connor Coley and Regina Barzilay of MIT, alongside Alexander Tropsha of the UNC Eshelman School of Pharmacy. According to the project plan, the models could produce billions of candidates. A second team will screen them, identifying molecules that hit their targets and feeding the resulting data back into the computational approach so it can improve.
From screening to greenhouse studies
Angela Koehler of MIT and Jaime Cheah of the Broad Institute will lead the high-throughput screening effort. After that process narrows the candidate pool, a third team will evaluate selected molecules in greenhouse studies under different agricultural conditions.
That group includes researchers from the University of Tennessee at Knoxville, the University of Illinois at Urbana-Champaign, the UNC Eshelman School of Pharmacy, and Robigo. The first greenhouse tests of initial RNAi candidates are expected to be complete by the end of the first year of the program.
Greenhouse testing is an important step in the project’s sequence: a candidate must move from a model prediction to a laboratory screen and then be evaluated in agricultural conditions. The article describes the research plan and expected early tests; it does not report that a new herbicide has already been developed or proven effective.
Reducing exposure is part of the plan
PLOW also includes development of living microbial systems intended to degrade agrochemicals before they reach food, soil, or waterways. The team plans to use engineered bacteria, fungi, and srRNA to convert chemicals and formulation ingredients into biomass.
Taken together, the work addresses weed control at two points: finding alternative molecules and reducing the presence of agrochemicals in the environment. The stated aim is to support farmers while reducing potential human and ecological exposure, though the project’s results will depend on the research and testing still to come.