AI Tool Accelerates Cryo-ET Membrane Analysis

Locating a single membrane protein inside a three-dimensional image of a cell has traditionally meant weeks of manual labeling. A new AI system built by researchers at Helmholtz Munich, the Technical University of Munich, and the Biozentrum of the University of Basel compresses that work into a few hours instead, according to a paper in Nature Methods describing the tool, called MemBrain v2.

The software targets a specific bottleneck in cryo-electron tomography, or cryo-ET, a technique that flash-freezes cells to preserve their structure and then images them in three dimensions at very high resolution. That freezing keeps everything from whole organelles down to individual molecules essentially intact, but the imaging process itself leaves gaps: certain angles of a membrane simply don’t show up well in the resulting data. “One challenge is that cryo-ET images can contain gaps in information due to technical limitations of the imaging process. As a result, certain membrane orientations are difficult or partly impossible to see. This is exactly where MemBrain v2 comes in, automating the process,” says first author Lorenz Lamm.

Rather than treating membrane detection, protein localization, and spatial analysis as three separate manual tasks, as earlier software required, MemBrain v2 runs all three through one pipeline: MemBrain-seg for finding membranes, MemBrain-pick for locating the proteins sitting within them, and MemBrain-stats for measuring how those proteins are arranged relative to each other. The segmentation module works without any extra annotation from the user. The protein-picking module needs barely more: given manual annotations for protein complexes on just one membrane in a test dataset, it went on to identify the same complexes on other membranes with an F1 score of 91 percent. Because the code is open source, the membrane-detection component alone has already spread well beyond its original developers, turning up in analyses of datasets from the Chan Zuckerberg Imaging Institute, among others. 

That combination of speed and low data requirements is the point, according to senior author Tingying Peng: “By making these analyses faster and accessible to research groups worldwide, we can study cellular processes across much larger datasets. This can ultimately help us better understand how cells function—and what changes when disease develops.”

The tool’s real-world payoff is already visible. A separate study built on MemBrain v2 found that certain photosynthesis proteins sit apart from each other within the membrane, an arrangement that runs counter to earlier models of how those proteins are organized. The developers are now working to sharpen the software’s ability to tell different protein types apart. “I’m especially pleased that MemBrain v2 is now being used in many further studies, where it simplifies demanding analyses or makes them possible in the first place—making a concrete contribution to new biological insights,” Lamm says.