AI will run experiments remotely in new Caltech lab

A $17 million grant from the National Science Foundation will fund a Caltech team led by professor Hosea Nelson to build ELECTRA, an automated and remotely accessible laboratory. The facility will employ microcrystal electron diffraction and artificial intelligence to determine the 3D structure of millions of molecules, structures Nelson refers to as mirroring the vast unknowns of dark matter in the universe.

“ELECTRA represents a bold move in accelerating scientific discovery,” says Caltech President Ray Jayawardhana, envisioning a resource that will fuel breakthroughs across multiple scientific fields. The project relies on the AI expertise of Caltech co-investigators Katie Bouman and Yisong Yue to automate data analysis and storage.

AI-Driven ELECTRA Lab to Tackle Chemical Dark Matter

Millions of molecules remain structurally unknown, a challenge Caltech professor Hosea Nelson terms as mirroring the vastness of undiscovered matter in the universe. Nelson’s team is addressing this gap with ELECTRA, an automated lab funded by a $17 million National Science Foundation grant, designed for remote experimentation and data analysis.

The facility will employ microcrystal electron diffraction (microED), a technique Nelson’s group pioneered in chemistry, adapting it from structural biology. MicroED excels at resolving atomic arrangements even with minuscule sample sizes, a critical advantage when dealing with rare natural products or compounds present in limited quantities.

This method has already yielded insights into molecules with potential pharmaceutical applications and foundational biological information, but scaling up the process demands automation and artificial intelligence. “We have tools that allow us to make guesses at many structures, but this project is about revealing all of those 3D structures on a large scale,” Nelson explains, framing the endeavor as similar to the Human Genome Project for natural products.

The cloud-based architecture of ELECTRA is central to its ambitious goals. The lab will house robotic systems for high-throughput experimentation and provide a platform for researchers nationwide to submit samples and access collected data. As we figure out what everything is, that will unlock new drugs, for example, and also advance basic biology, where mysteries often arise from a lack of knowledge about molecular structures, behaviors, and functions.

There are millions of such structures yet to be solved, and the tool is useful in a wide variety of disciplines. The team also includes Jose A. Rodriguez of UCLA, Garret Miyake of Colorado State University, Fort Collins, Emily Balskus of Harvard University, and Alison Narayan of the University of Michigan.

Biomedical applications are one area; for example, taxol, used for chemotherapy, is derived from the Pacific yew tree, and acetylsalicylic acid, the active ingredient in aspirin, originally came from the bark of willow trees. The team anticipates a dramatic reduction in the time required to determine molecular structures.

What once took decades, from the 1920s, 1940s, 1990s, and 2018, could potentially be accomplished in under a minute. “In the 1920s, figuring that out meant a Nobel Prize,” Nelson remarked, referencing the historical significance of structural elucidation. “Now we want to make that kind of determination in less than a minute.” This leap in efficiency is not merely about speed; it’s about unlocking a vast reservoir of chemical knowledge and accelerating innovation across a wide range of scientific fields.

NSF Funds Caltech Cloud Laboratory for Autonomous Experimentation

Over the past decade, Nelson’s team has successfully employed microED to decipher the structures of numerous small molecules derived from natural sources, providing crucial data for both synthetic chemists and biologists. However, the sheer scale of unsolved molecular structures necessitates a dramatically accelerated approach, prompting the development of this automated, cloud-based system.

This data repository will serve as a training ground for more sophisticated AI models, enabling the prediction of molecular structures from limited data, and ultimately accelerating the pace of scientific discovery across multiple disciplines. The scope of ELECTRA extends beyond fundamental chemical research, encompassing applications in biomedicine, geology, and materials science.

Nelson highlights the potential for discovering novel pharmaceuticals derived from unexplored natural products, giving examples like taxol and acetylsalicylic acid. The lab will support investigations into the complex molecular interactions within biological systems, potentially unlocking new insights into disease mechanisms and therapeutic targets. The collaborative nature of the ELECTRA project is noteworthy, bringing together experts from diverse fields.

Jose A. Rodriguez of UCLA brings structural biology expertise to the team. Nelson explains the rationale behind this interdisciplinary approach, stating, “We don’t know what these molecules are in many cases, so we pulled together experts in disparate fields to help us build out a platform that can apply to all of them.” He draws a parallel to the Human Genome Project, suggesting that ELECTRA aims to achieve a similar level of comprehensive structural characterization for the vast world of natural products.

A significant challenge lies in the limited availability of data for training the AI algorithms. Unlike fields like protein structure prediction, where large datasets already exist, chemistry lacks a comparable resource.