MIT-led plan for weed-killing AI-designed herbicide candidates gets US backing

A team led by the Massachusetts Institute of Technology will receive up to $53 million from the U.S. Advanced Research Projects Agency for Health (ARPA-H) to develop AI-designed herbicides that work against resistant weeds and pose less risk to people. The work relies on generative AI, software that learns patterns in existing data and proposes new candidates, in this case molecules.

The project is called Precision Learning for On-target Weed control (PLOW). Christopher Voigt, head of MIT’s Department of Biological Engineering, is the principal investigator. Voigt said glyphosate and atrazine face “a real crisis because of resistance mutations that are coming.”

More than 1 billion pounds (about 454 million kilograms) of herbicide are used in the U.S. each year, MIT said, and some weeds are becoming resistant to the most common products. MIT said exposure may contribute to cancer and metabolic and endocrine disorders, and that small amounts reach food and drinking water.

A three-team search for new compounds

The award falls under ARPA-H’s Systems for Tracking and Resilience in Efficient Agricultural-input Management (STREAM) program, which the Department of Health and Human Services launched Aug. 17 to find safer weed-control tools. MIT’s announcement says three teams will split the work.

A generative AI group led by Connor Coley, Regina Barzilay and Alexander Tropsha will build models that predict compounds, or small interfering RNA (siRNA, short genetic fragments that silence a specific gene), able to disrupt the plant enzymes that glyphosate and other herbicides target. Glyphosate blocks EPSP synthase, an enzyme plants need to make several amino acids.

The models will also weigh synthesis cost, resistance risk and potential harm to humans or ecosystems. They are expected to generate billions of candidates. A second group, led by Angela Koehler and Jaime Cheah, will screen them and send results back. Voigt said the screening data will go to the computational approach so it can improve. MIT researchers have also released separate AI molecule screening tools for drug discovery.

Greenhouse tests and open questions

A third team will test the narrowed candidates in greenhouses under varied farm conditions. It is led by Neal Stewart and Scott Lenaghan of the University of Tennessee at Knoxville, Patrick Tranel of the University of Illinois at Urbana-Champaign, Tim Willson of the University of North Carolina at Chapel Hill, and Andee Wallace of Robigo.

The project also includes engineered bacteria and fungi meant to convert agrochemicals into biomass before they reach food, soil or waterways. Separately, a University of British Columbia fruit wash removes 86% to 94% of pesticide residue from apples in lab tests.

Nothing is proven yet. MIT describes plans and not tested molecules, and gives no performance data. The $53 million is a ceiling. MIT also said the alternatives could cut U.S. reliance on imported phosphorus, a key ingredient in glyphosate, but gave no figures.

Greenhouse results for the first RNA-based candidates are due by the end of the first year.

Source: https://interestingengineering.com/ai-robotics/ai-designed-herbicides-mit-arpa-h