Innovation, thanks to AI and deep learning it is possible to observe protein aggregation

Using an autonomous imaging system based on deep learning and artificial intelligence , it was possible to observe the aggregation of misfolded proteins. Scientists at the Swiss Federal Institute of Technology in Lausanne (EPFL) have achieved this feat, publishing a paper in the journal Nature Communications to present their findings. The team, led by Khalid Ibrahim, Aleksandra Radenovic, Hilal Lashuel, and Robert Prevedel, developed an autonomous imaging system that uses various microscopy methods to track and analyze protein aggregation in real time, or anticipate it before it begins. This approach, the experts reveal, maximizes imaging efficiency and minimizes the use of fluorescent markers, which can alter the biophysical properties of cell samples and hinder the accuracy of the analysis. "For the first time," says Ibrahim, "we have been able to accurately predict the formation of these protein aggregates. Understanding how these properties evolve during the aggregation process will lead to a fundamental understanding, essential for developing solutions." The research team has developed a deep learning algorithm capable of detecting mature protein aggregates.

In the new work, the researchers developed a new version of the algorithm that analyzes images in real time. If a mature protein aggregate is detected, a Brillouin microscope is activated, analyzing the scattered light to characterize the biomechanical properties of the aggregates, such as elasticity. Thanks to artificial intelligence, the microscope is activated only if a protein aggregate is detected , accelerating the process and paving the way for intelligent microscopy. To capture the ongoing formation of aggregates, the researchers developed a second algorithm, trained on fluorescently labeled images of proteins in living cells. This system allows the system to distinguish between nearly identical images and correctly identify the moment at which aggregation will occur with 91 percent accuracy. This work, the authors comment, has important implications for drug discovery and precision medicine. “Label-free imaging approaches,” Lashuel concludes, “create entirely new ways to study and target small protein aggregates called toxic oligomers, which are thought to play a central causal role in neurodegeneration.”