The buildup of misfolded proteins in the brain plays a key role in diseases such as Huntington’s, Alzheimer’s, and Parkinson’s. However, these harmful proteins appear indistinguishable from normal ones to the human eye.
The formation of protein aggregates typically occurs unpredictably and within minutes—making detection and analysis difficult. Yet, identifying and understanding these aggregates is crucial to advancing treatments for neurodegenerative diseases.
AI-Driven Microscope Predicts Protein Aggregation in Real Time Without Disruptive Tags
Now, researchers at EPFL have developed a deep learning–powered, “self-driving” microscope system that integrates multiple imaging techniques to monitor protein aggregation in real time—and even predict it before it begins. This method improves imaging efficiency while minimizing the use of fluorescent tags, which can interfere with a sample’s natural behavior and reduce accuracy.
“This is the first time we’ve been able to reliably anticipate the formation of protein aggregates,” said Khalid Ibrahim, a recent EPFL Ph.D. graduate. “Since their mechanical properties are tied to disease and cell dysfunction, understanding how these evolve during aggregation is key to developing effective treatments.”
The research, published in Nature Communications, was led by Ibrahim alongside Aleksandra Radenovic of EPFL’s Laboratory of Nanoscale Biology and Hilal Lashuel from the School of Life Sciences, in partnership with Carlo Bevilacqua and Robert Prevedel of the European Molecular Biology Laboratory in Heidelberg.
The work stems from a long-term collaboration between Lashuel’s and Radenovic’s labs, combining expertise in neurodegenerative disease and advanced live-cell imaging.
“This project began with a desire to develop methods that uncover new biophysical insights,” said Radenovic. “It’s incredibly rewarding to see that vision realized.”
In their initial collaboration, led by Khalid Ibrahim, the researchers created a deep learning algorithm capable of detecting mature protein aggregates in unlabeled images of live cells.
Enhanced Algorithm Triggers Real-Time Brillouin Imaging to Reveal Protein Aggregate Mechanics
Building on that foundation, the new study introduces an upgraded version of the algorithm designed for real-time image classification. When this algorithm identifies a mature aggregate, it activates a Brillouin microscope, which uses scattered light to measure the aggregates’ biomechanical properties—such as elasticity.
Typically, Brillouin microscopy is too slow for tracking fast-forming aggregates. However, the AI-enhanced approach ensures the microscope only runs when needed, significantly improving efficiency and paving the way for more adaptive, intelligent microscopy.
“This is the first study to demonstrate how self-driving systems can successfully integrate label-free microscopy, making it easier for biologists to adopt advanced smart imaging tools,” said Ibrahim.
Since this algorithm only detects fully formed aggregates, the team took an extra step to capture the early stages of aggregation. They developed a second deep learning model, trained on fluorescently tagged images of live-cell proteins, to predict aggregation before it occurs.
The newly developed “aggregation-onset” detection algorithm can distinguish between nearly identical images to accurately predict when protein aggregation is about to begin—achieving 91% accuracy. Once the system detects the early stages, it activates the Brillouin microscope, offering an unprecedented real-time view into the biomechanics of aggregation as it unfolds.
According to Lashuel, beyond advancing smart microscopy, the technology holds significant promise for drug development and precision medicine.
“Label-free imaging offers entirely new opportunities to investigate and target small, toxic protein clusters known as oligomers, which are believed to be key drivers of neurodegenerative diseases,” he explains.
“We’re excited to expand on these results and lay the foundation for drug discovery platforms that can fast-track the development of more effective treatments.”
Read the original article on: Phys Org
Read more: Converting Plastic to Gasoline: Innovative Science or Just TikTok Trend?