Two researchers at the Department of Energy's SLAC National Accelerator Laboratory have been awarded $550,000 each over five years to develop artificial intelligence tools that unlock hidden information about how proteins move and how battery materials behave at the molecular level. The awards, granted by the DOE Office of Science Early Career Research Program, represent a significant investment in using machine learning to speed up scientific discovery in drug development and energy storage, two critical areas where traditional trial-and-error methods have slowed progress for decades.

Why Do Proteins Need AI to Reveal Their Secrets?

Proteins are often described as "dynamic molecular machines," constantly bending, rotating, and changing shape as they interact with other molecules. This movement is crucial for understanding how drugs bind to disease-causing proteins, yet most X-ray crystallography experiments capture only static snapshots of protein structure, like a single frame from a video rather than the full motion. Derek Mendez, a staff scientist in the Structural Molecular Biology division at SLAC's Stanford Synchrotron Radiation Lightsource (SSRL), is developing software to extract this hidden dynamic information from crystallography data.

"Proteins are commonly referred to as dynamic molecular machines. Protein structures can bend or rotate, thus changing how they might bind with other molecules, so it's critical to understand how they move and change shape," said Mendez.

Derek Mendez, Staff Scientist, Structural Molecular Biology Division at SSRL

Mendez's approach combines two powerful techniques. First, he will use a machine learning method called multi-objective Bayesian optimization to extract dynamics information already hidden in existing crystallography data. Second, he plans to develop AI models trained on simulated data that can suggest experimental conditions most likely to reveal protein movement or drug binding, an approach called AI-assisted experimental steering. His initial focus targets proteins involved in antibiotic resistance, with plans to extend the work to other applications in medicine, bioenergy, and biopreparedness.

How Can AI Accelerate Battery and Energy Storage Research?

The second award recipient, Xueli "Sherry" Zheng, is tackling a different but equally complex challenge: understanding the chemistry at the interfaces inside battery systems. Battery performance depends heavily on what happens at the boundary between the cathode, electrolyte, and anode, where a protective layer called the cathode-electrolyte interphase (CEI) forms. This interface chemistry is so complex that researchers have struggled to understand and control it, making it one of the grand challenges in energy storage.

Zheng's program will combine machine learning models with state-of-the-art experimental techniques, including X-ray spectroscopy, cryogenic electron microscopy, and tomography conducted under realistic operating conditions. Her vision is to build a machine learning model that predicts and guides the design of optimal cathode-electrolyte interfaces for sodium-ion battery systems, which are emerging as a more sustainable alternative to lithium-ion batteries.

"One of the most important factors to influence the system's performance is the electrode-electrolyte interface. Understanding and controlling the chemical dynamics at this interface remains one of the grand challenge in this field," explained Zheng.

Xueli Zheng, Staff Scientist, SLAC-Stanford Battery Center

How to Leverage AI and Experimental Data for Materials Discovery

• Combine Physics Models with Data-Driven AI: Rather than relying on either theoretical models or experimental data alone, integrating both allows AI systems to learn from theory, fill in knowledge gaps, and identify optimal conditions faster than manual methods.
• Use Real-Time Sensors During Experiments: Embedding sensors directly into experimental systems generates rich, continuous data about chemical reactions as they occur, reducing the need for bulky external instruments that can disrupt reactions.
• Apply Machine Learning to Suggest Next Experiments: AI models trained on simulated or historical data can predict which experimental conditions are most likely to reveal important information, reducing the number of trial-and-error attempts needed to optimize materials or drug candidates.

Both researchers emphasize that their work represents a collaboration between human expertise and artificial intelligence. Mendez noted that he looks forward to "combining human and artificial intelligence to tackle hard problems in the study of protein dynamics." Similarly, Zheng's approach builds on the unique capabilities of SLAC's world-class facilities, including synchrotron radiation sources and cryogenic electron microscopy centers, paired with computational infrastructure.

Mendez

The broader impact of these awards extends beyond the specific applications. Similar AI-assisted experimental approaches are being explored in other fields, including carbon dioxide conversion research, where miniature lab-on-a-chip devices equipped with real-time sensors and AI optimization are accelerating the discovery of catalysts for sustainable fuels and chemicals. This convergence of experimental hardware, sensor technology, and machine learning represents a shift in how materials science research is conducted, moving from qualitative observation toward quantitative, high-throughput data collection that can be rapidly analyzed and refined.

The DOE Early Career Research Program is designed to support promising researchers in the early stages of their careers as they develop innovative approaches to solving scientific challenges. By investing in Mendez and Zheng's work, the DOE is betting that AI-enhanced experimental methods will significantly compress the timeline from material discovery to practical applications in drug development and clean energy technologies, both critical priorities for national competitiveness and addressing climate change.