Particles of different macromolecules are arranged in the map according to their structure, allowing users to identify and locate different macromolecules in cells.
Electron cryo-tomography (cryo-ET) is emerging as a powerful technique for providing detailed 3D images of the cellular environment and captured biomolecules. However, one of the challenges of the methodology is the identification of protein molecules in the images for further processing.
A research team led by Stefan Raunser, MPI Director of Molecular Physiology in Dortmund, and led by Thorsten Wagner, developed software to select proteins in crowded cell volumes.
A new open-source tool called TomoTwin is based on deep metric learning and allows scientists to localize multiple proteins with high accuracy and throughput without having to manually build or retrain the network each time.
“TomoTwin paves the way for the automated identification and localization of proteins directly in their cellular environment, expanding the potential of cryo-ET,” says Gavin Rice, co-author of the paper. Cryo-ET has the potential to reveal the workings of biomolecules in the cell and reveal the basis of life and the origin of disease.
In the cryo-ET experiment, scientists use a transmission electron microscope to obtain 3D images, called tomograms, of a volume of cells containing complex biomolecules. To get a more detailed picture of each protein, they average as many copies of it as possible like photographers who take the same photo with different exposures to later combine them into a perfectly exposed image.
Crucially, it is possible to correctly identify and localize the different proteins in the image before averaging them. “Scientists can create hundreds of tomograms a day, but we haven’t had the tools to fully identify the molecules they contain,” says Rice.
Until now, researchers have used algorithms based on templates of already known molecular structures to find matches in tomograms, but these tend to be error-prone. Manual identification of molecules is another option that provides high-quality selection, but takes days to weeks per data set.
Another option would be to use some form of supervised machine learning. These tools can be very accurate, but are currently not user-friendly, as they require manual labeling of thousands of samples to train the software on each new protein, an almost impossible task for small biological molecules in a crowded cellular environment.
The newly developed TomoTwin software overcomes many of these obstacles: it learns to pick out molecules that have a similar shape in the tomogram and maps them into geometric space—the system is rewarded for placing similar proteins close together, or penalized. The new map allows researchers to isolate and precisely identify different proteins and thus localize them in the cell.
“One advantage of TomoTwin is that we provide a pre-trained picking model,” says Rice. By removing the training step, the software can run even on local computers – where a tomogram typically takes 60-90 minutes to process, the run time on the Raven MPI supercomputer is reduced to 15 minutes per tomogram.
TomoTwin allows researchers to select dozens of tomograms in the time it takes to manually select a single one, increasing data throughput and averaging speed to obtain a better image. The software can currently localize globular proteins or protein complexes larger than 150 kilodaltons in cells; In the future, Raunser’s group would like to include membrane proteins, filamentous proteins, and proteins of smaller size.
Read Now:NASA’s Webb finds water and new mystery in rare main belt comet
