Cracking the Code: Using AI to Solve Difficult-to-Map Proteins

Using a tool to solve a protein’s structure, for most researchers in the world of structural biology and computational chemistry, is not unlike using the Rosetta Stone to unlock the secrets of ancient Egyptian texts. Once a protein’s structure has been discovered, or defined, one can infer crucial information about its function or, in a diseased state, its dysfunction. While researchers have been pursuing the quest of solving protein structure for decades, advancing tools and computing technologies offer a new frontier for this work.

A collaborative study recently published in Nature Communications unveiled a new computing program that offers a faster and more accurate way to determine protein structure at a new level of precision. Researchers from the Department of Energy’s Lawrence Berkeley National Laboratory (Berkeley Lab), along with an international team of researchers, were a part of the effort. This tool, dubbed AI-enabled Quantum Refinement, or AQuaRef for short, uses quantum-mechanical calculations (QM) and artificial intelligence (AI) to predict the highly-accurate placement of atoms and electrons to determine a protein’s molecular structure.

This program is a part of Phenix, a comprehensive software suite that generates realistic computer models used by structural biologists around the world to solve macromolecular structures. “We’re all basically a bunch of proteins,” said Nigel Moriarty, a Berkeley Lab researcher and contributor to the recent publication. “They do so much in our bodies that detail the processes of life. Understanding their structure can give us insights into the mechanisms that cause disease in humans or produce energy in plants. All of this knowledge can lead to more effective therapeutics and bioenergy production.”

“We’re all basically a bunch of proteins. They do so much in our bodies that detail the processes of life. Understanding their structure can give us insights into the mechanisms that cause disease in humans or produce energy in plants. All of this knowledge can lead to more effective therapeutics and bioenergy production.” — Nigel Moriarty

The current way of mapping a protein’s structure entails bringing together two streams of information: experimental data produced through techniques like X-ray crystallography and cryogenic electron microscopy (cryo-EM), and theoretical data that exists in a library of detailed, known protein structural information. But the current options are limited, explained Moriarty, a computational research scientist in the Molecular Biophysics and Integrated Bioimaging (MBIB) Division’s Phenix group. Our understanding today is limited to the chemical entities that have already been defined and doesn’t yet include meaningful noncovalent interactions, the type of attraction typically seen holding a protein in its structural form. “That’s where quantum and AI come in,” he said.

Nearly five years ago, members of the Phenix team began working with researchers at Carnegie Mellon University to explore how they might be able to apply their coding work to Phenix’s offerings. The collaborative approach, coupled with 15 years of incremental research, led to this breakthrough program. In addition to Moriarty, other members of the Phenix team involved in this work were Paul Adams and Billy Poon, with Pavel Afonine leading the research. AQuaRef uses machine learning (ML) tools developed at Carnegie Mellon integrated with the Phenix software to compute energy and forces for scientifically interesting proteins—making quantum-level refinement practical where it was previously impossible.

Of the 71 experiments that were tested in this study, AQuaRef produced higher quality structural information at a substantially lower computational cost while maintaining an equal or better fit to experimental data. In addition to the proof-of-concept results from this work, AQuaRef also correctly determined proton positions in DJ-1, a human protein linked to some forms of Parkinson’s Disease, the structure of which has been notoriously difficult to map. Now that the team has confirmed that quantum-level refinement of a 3D protein model structure is possible, they’re aiming to broaden the scope to include more diverse structures, such as those required for pharmaceutical drug design. And the potential impacts of this work reach far beyond human health, from better understanding the mechanisms of photosynthesis for enhanced crop productivity to mapping the proteins in plants as it relates to biofuel production.

“There is a near-infinite number of things that can benefit from a detailed understanding of these mechanisms and protein structure,” said Moriarty. “I’m excited to see how the paradigm shift that AQuaRef represents impacts the field of protein structure determination.”

This international team also included collaborators from the University of Wrocław, Poland, the University of Florida, and Pending.AI, Australia.

This work was funded by the National Institutes of Health as well as with support from the Phenix Industrial Consortium.

###

Lawrence Berkeley National Laboratory (Berkeley Lab) is committed to groundbreaking research focused on discovery science and solutions for abundant and reliable energy supplies. The lab’s expertise spans materials, chemistry, physics, biology, earth and environmental science, mathematics, and computing. Researchers from around the world rely on the lab’s world-class scientific facilities for their own pioneering research. Founded in 1931 on the belief that the biggest problems are best addressed by teams, Berkeley Lab and its scientists have been recognized with 17 Nobel Prizes. Berkeley Lab is a multiprogram national laboratory managed by the University of California for the U.S. Department of Energy’s Office of Science.

DOE’s Office of Science is the single largest supporter of basic research in the physical sciences in the United States, and is working to address some of the most pressing challenges of our time. For more information, please visit energy.gov/science.