AI Predicts Protein-Metal Interactions to Boost Biofuel Crop Resilience and Sustainability

Scientists Develop AI to Predict Protein Structures and Functions

Researchers have created an artificial intelligence (AI) workflow that predicts the 3D structures and functions of unknown proteins, including their interactions with metals like zinc. For instance, the model identified a zinc-binding protein where four cysteine residues are key in binding zinc.

New AI Model for Protein Prediction

At the U.S. Department of Energy’s Brookhaven National Laboratory, biologists and computational scientists refined two AI programs originally developed by Meta (Facebook’s parent company). The combined model, named ESMBind, predicts protein shapes and reveals how they bind to vital metals such as zinc and iron. These metals are essential for biological processes, and understanding their interaction with proteins helps researchers explore how plants absorb nutrients from soil.

One practical goal is to engineer biofuel crops that thrive in nutrient-poor soils, preserving fertile land for food production. As Qun Liu, a Brookhaven Lab structural biologist, stated, “We do not want biofuel crops to compete with crops for food. Instead, we need to grow these bioenergy plants on nutritionally deficient land.”

Proteins Bind to Metals Necessary for Life

Proteins are chains of amino acids that fold into specific 3D shapes. This folding brings certain amino acids close together, defining how proteins interact with other molecules. ESMBind predicts these shapes to infer protein functions, especially their metal-binding capabilities.

“We believe there’s opportunity to leverage machine learning, a form of AI, to speed up the creation of useful protein models,” Liu explained. The ESMBind model can run hundreds of thousands of simulations daily. AI scientist Xin Dai and his team started with two Meta foundation models—ESM-IF and ESM-2—to extract information from protein sequences and structures. Their combined workflow predicts whether a protein can bind a specific metal.

Traditionally, scientists determine protein structures experimentally using facilities like the National Synchrotron Light Source II (NSLS-II), which employs ultra-bright X-rays to reveal atomic details. Much of the data training ESMBind came from X-ray crystallography at NSLS-II and other synchrotrons. However, such experiments are time-consuming. ESMBind serves as a screening tool to identify promising proteins, reducing the experimental workload. Compared to other AI models, ESMBind showed superior accuracy in predicting protein structures and their functions.

Applying AI to Bioenergy Research

The team is focusing on sorghum, a crop well-suited for bioenergy due to its ability to grow on marginal lands with limited nutrients and tolerate heat. Sorghum can be converted into biofuels like ethanol and biochar. Understanding how sorghum proteins interact with soil metals could improve its bioenergy applications.

Additionally, the researchers applied ESMBind to proteins from Colletotrichum sublineola, a fungus that infects sorghum. Like plant proteins, fungal proteins bind metals that facilitate infection. By modeling these metal-binding sites, they aim to find ways to disrupt fungal infectivity and protect sorghum. The study identified about 140 candidate fungal proteins that may contribute to infection, providing targets for future disease control strategies.

“Protecting plants and biofuel crops from infectious diseases is a research priority for the plant sciences group within the Brookhaven Lab Biology Department,” Liu added.

Future Directions

Looking ahead, the scientists plan to enhance the ESM-based model to design proteins capable of extracting and separating critical minerals and materials from sources like mine tailings and ores. Current extraction methods use harsh chemicals and consume significant energy. Proteins engineered to capture specific minerals, including rare earth elements, could provide a more sustainable alternative.

“If we can design a protein to fold and capture a rare earth element in a specific way, we might be able to engineer microbes to produce that protein and use them to extract and recover that critical mineral,” Liu explained.

ESMBind is open source and accessible to the research community for generating protein-metal interaction models. This project was supported by Brookhaven Lab’s Laboratory Directed Research and Development program and the DOE Office of Science, which is the largest U.S. supporter of basic physical sciences research addressing key scientific challenges.