AI Tool Cuts Weeks of 3D Cell-Membrane Mapping Down to a Few Hours

A team from Helmholtz Munich, the Technical University of Munich (TUM) and the Biozentrum of the University of Basel has built an AI tool that automates the analysis of cell membranes in 3D microscope images, cutting work that once took weeks down to a few hours. The tool, called MemBrain v2, is described in a study published in Nature Methods.

Cell membranes and the proteins within them control many processes important to health and disease, but studying them in three-dimensional images has so far meant slow, manual work. MemBrain v2 analyses data from cryo-electron tomography (cryo-ET), a microscopy technique that flash-freezes cells to preserve their structure, letting researchers see inside cells in 3D at very high resolution. Cryo-ET images can contain gaps from technical limits of the imaging process, so some membrane orientations have been difficult to see, said first author Lorenz Lamm.

The free, open-source software combines three steps previously handled by separate, single-purpose programs: MemBrain-seg finds membranes directly without extra annotations, MemBrain-pick locates the proteins embedded in them, and MemBrain-stats measures how those proteins are spatially arranged. In one test, researchers manually annotated protein-complex positions on just a single membrane, and MemBrain-pick then localised the same complexes on additional membranes with an F1 score of 91%.

"By making these analyses faster and accessible to research groups worldwide, we can study cellular processes across much larger data sets," said senior author Dr. Tingying Peng. "This can ultimately help us better understand how cells function — and what changes when disease develops." The membrane-detection module is already used worldwide, including on data sets from the Chan Zuckerberg Imaging Institute.

MemBrain v2 has already contributed to new biology: in a separate study, the tool showed that key photosynthesis proteins are spatially separated within the membrane, challenging earlier models of how they are organised. "I'm especially pleased that MemBrain v2 is now being used in many further studies," Lamm said, adding that the tool is set to distinguish different protein types even more precisely in the future.