A new study from USDA’s Agricultural Research Service (ARS) and Iowa State University (ISU) reveals that generative Artificial Intelligence (AI) can help expedite the search for solutions to reduce enteric methane emissions caused by cows in animal agriculture, which accounts for about 33 percent of U.S. agriculture and 3 percent of total U.S. greenhouse gas emissions.
“Developing solutions to address methane emissions from animal agriculture is a critical priority. Our scientists continue to use innovative and data-driven strategies to help cattle producers achieve emission reduction goals that will safeguard the environment and promote a more sustainable future for agriculture,” said ARS Administrator Simon Liu.
One of these innovative solutions starts in the cow’s stomach, where microorganisms contribute to enteric fermentation and cause cows to belch methane as part of normal digestion processes. The team of scientists found a group of compound molecules capable of inhibiting methane production in the largest of the cow’s four stomach compartments, the rumen, which can be tested to help mitigate methane emissions.
One molecule in particular, bromoform, which is naturally found in seaweed, has been identified by the scientific community to demonstrate properties that can result in reducing cattle enteric methane production by 80-98 percent when fed to cattle. Unfortunately, bromoform is known to be a carcinogen, limiting its potential use in cattle for food safety reasons. Therefore, scientists continue to search for molecules with similar potential to inhibit enteric methane. However, this type of research presents challenges of being especially time-consuming and expensive.
In response to these challenges, a team of scientists at the ARS Livestock Nutrient Management Research Unit and ISU’s Department of Chemical and Biological Engineering combined generative AI with large computational models to jumpstart the quest for bromoform-like molecules that can do the same job without toxicity.
“We are using advanced molecular simulations and AI to identify novel methane inhibitors based on the properties of previously investigated inhibitors [like bromoform], but that are safe, scalable, and have a large potential to inhibit methane emissions,” said Matthew Beck, a research animal scientist working with ARS at the time the study was completed and is now with Texas A&M University’s Department of Animal Science. “Iowa State University is leading the computer simulation and AI work, while ARS is taking the lead in identifying compounds and truth testing them using a combination of in vitro [laboratory] and in vivo [live cattle] studies.”
Publicly available databases that contained scientific data collected from previous studies on the cows’ rumen were used to build large computational models. AI, along with these models, was used to predict the behavior of molecules and to identify those that can be further tested in a laboratory. The results from the laboratory tests feed the computer models for AI to make more accurate predictions, creating a feedback loop process known as a graph neural network.
“Our graph neural network is a machine learning model, which learns the properties of molecules, including details of the atoms and the chemical bonds that hold them, while retaining useful information about the molecules’ properties to help us study how they are likely to behave in the cow’s stomach,” said ISU Assistant Professor Ratul Chowdhury. “We studied their biochemical fingerprint to identify what makes them do the job successfully as opposed to the other fifty thousand molecules that are lurking around in the cow’s rumen but don’t actively stop the production of methane.”
“This study successfully demonstrated that fifteen molecules cluster very close to each other in what we call a ‘functional methanogenesis inhibition space,’ meaning they seem to contain the same enteric methane inhibition potential, chemical similarity, and cell permeability as bromoform,” added Chowdhury.
Scientists believe AI can play a significant role in understanding how known molecules interact with both proteins and the microbial community of the rumen and thereby discover novel molecules and potentially key interactions within the rumen microbiome. This type of predictive modeling can be particularly helpful for animal nutritionists.
“There are other promising strategies currently available to mitigate enteric methane emissions, but the available solutions are relatively limited,” said USDA-ARS Research Leader Jacek Koziel. “This is why combining AI with laboratory research, through iterative refinement, is a valuable scientific tool. AI can fast-forward the research and accelerate these several pathways that animal nutritionists, researchers, and companies can pursue to get us closer to a very ambitious goal of limiting greenhouse gas emissions and helping mitigate climate change.”
The study also presents a total computational and monetary cost breakdown to conduct this research on a per molecule basis. This analysis was conducted to show an estimate of potential costs and foreseeable pitfalls of this research. This estimate can be used to guide decision-making on investments for this type of research to be done entirely in a laboratory.
Chowdhury, Beck, and Koziel are co-authors in the paper published in Animal Frontiers, along with Nathan Frazier (ARS) and Logan Thompson (Kansas State University). Mohammed Sakib Noor, an ISU graduate student, is working with Chowdhury to develop the graph neural networks.
The Agricultural Research Service is the U.S. Department of Agriculture’s chief scientific in-house research agency. Daily, ARS focuses on solutions to agricultural problems affecting America. Each dollar invested in U.S. agricultural research results in $20 of economic impact.