KEY TAKEAWAYS
- HHMI Investigators Eric Gouaux and Michael Rosen are spearheading the development of advanced AI tools to bolster cryo-electron tomography (cryo-ET), which is a microscopy technique capable of imaging cells and complex biochemical arrangements in three dimensions.
- Their efforts integrate AI with innovative methods such as gold nanoparticle labeling to enhance the interpretation of cryo-ET images, facilitating the identification of molecules within larger structures, including tightly coiled DNA strands and neuronal communication sites.
- Grasping the spatial organization of these molecules can provide insights into the functioning of the larger structures they compose and the implications of their dysfunction.
- This project is part of AI@HHMIexternal link, opens in a new tab, an ambitious $500 million initiative aimed at advancing AI-driven projects and integrating AI systems throughout the scientific landscape.
While biologists often embrace simplicity, incorporating a touch of sophistication can greatly enhance cryo-electron tomography.
Cryo-electron tomography (cryo-ET) offers a remarkable 3D visualization of cells within their natural settings, delivering high-resolution insights into biological structures. Although adept at unveiling sizable components, like cellular organelles, it poses challenges for discerning the individual molecules that constitute these structures—crucial knowledge for understanding their functions.
To tackle this issue, researchers apply tiny gold nanoparticles that bond to the specific molecules of interest. However, spotting these nanoparticles amidst the visual “noise” in the images can be both daunting and time-consuming.
HHMI Investigator Eric Gouaux compares the search to detectives Sherlock Holmes and Doctor Watson tracking down Professor Moriarty on a foggy London night—while they can spot Moriarty’s faint lantern, pinpointing the mastermind’s features remains a challenge.
But now, scientists have an ally in artificial intelligence.
Utilizing AI to Reveal Molecular Details in Cryo-ET
Gouaux, who heads a lab at Oregon Health and Science Universityexternal link, opens in a new tab, alongside Michael Rosen from UT Southwestern Medical Centerexternal link, opens in a new tab, is leading an initiative to leverage AI for enhancing cryo-ET. Their goal is to shift the focus from merely imaging large cellular structures to also examining the minute molecules that compose them.
Recognizing the organization of these molecules provides biologists insight into how larger structures operate and the effects of any dysfunctions.
For Rosen, studying how chromatin—highly compacted DNA—is organized inside the nucleus is essential in unlocking its role in processes such as gene expression. Similarly, analyzing the molecular arrangement in synapses, the connections where neuron signals are exchanged, aids Gouaux’s team in understanding how these transmissions occur.
“Ultimately, our goal is to discern whether Moriarty is armed and how we can determine that,” Gouaux reflects. “With respect to these assemblies, our aim is to intricately understand their connections and mechanics.”
Training AI Models to Identify Molecules
Utilizing cryo-ET data generated by the researchers, the team intends to develop and train AI models that can rapidly and accurately identify gold nanoparticle labels across varied shapes and brightness, enhancing the speed and precision far beyond human capability.
Additionally, the researchers aim to broaden the range of gold nanoparticle labels they employ and explore different labeling types compatible with other imaging methods. This could unveil further molecular details, according to HHMI Investigator Elizabeth Villa, a project collaborator.
Tomography is currently the most advanced imaging tool available to biologists, yet identifying specific molecules remains a challenge. Enhancing these labeling methods and simplifying their use would be revolutionary for researchers, Villa asserts.
“I can’t imagine a single researcher using tomography who wouldn’t desire a plug-and-play solution, nor can I think of many who wouldn’t want to dive into it, even if they’re not tomography experts,” she adds.
The team is also set to create and train AI models based on dense chromatin biochemical reconstitutions and molecular simulations developed by Rosen’s collaborator, Rosana Collepardo-Guevara at the University of Cambridge. Insights from both the biochemistry and simulations will refine the models’ predictions regarding chromatin locations within tomograms, connections among various elements, and the implications of changes like mutations, diseases, and aging on chromatin structure and function.
These AI-enhanced techniques will not only improve the quality of cryo-ET images—yielding more accurate reconstructions of molecular structures—but will also facilitate deeper analysis, allowing researchers to extract richer data, explains Magdalena Schneider, a Machine Learning Researcher in the AI@HHMI Initiative at HHMI’s Janelia Research Campusexternal link, opens in a new tab, who is collaborating on the project.
“We have the ability to observe phenomena that were previously hidden within the cell’s natural environment,” she remarks.
Elevating Tomography to New Horizons
Ultimately, the team aspires to create AI tools accessible to biologists in laboratories across the globe, enhancing the understanding of a myriad of molecular structures beyond just synapses and chromatin.
“If the AI can establish a generalizable method for labeling multiple objects distinctively, and quickly identify their locations and structures, it will be transformative across diverse fields,” Rosen notes.
The researchers acknowledge that the AI@HHMI Initiativeexternal link, opens in a new tab empowers them to achieve remarkable advances beyond what their individual labs could accomplish, facilitating collaboration with experts in AI, cryo-ET, and annotation, all centralized at HHMI’s Janelia Research Campus.
“This undertaking wouldn’t be feasible without this collaborative environment,” Gouaux concludes. “It’s genuinely transformative.”
