🟢 Microtubule

easymode segment microtubule

The microtubule model was trained to output a cylindrical tube with a diameter of 180 Å along microtubules, approximately annotating the lumen of microtubules only. As a result, closely adjacent microtubules are still segmented as individual filaments.

In combination with the --filament flag in easymode pick, the model enables tracing individual microtubules and picking particles at regular intervals along individual filaments, with an accurate prior on the particle orientation and class labels linking particles to the parent filaments (_aisFilamentID). Use the --per-filament flag to write .star files for individual microtubules. This allows subtomogram averaging of individual filaments.

For validation, we segmented, picked, and averaged microtubules in a large dataset of FIB-milled HeLa cell tomograms. After using per-filament averaging to determine polarity and protofilament count, then subboxing individual protofilaments, we achieved a 4.6 Å overall resolution (49k particles).

Note

At 4.6 Å resolution, differences between alpha and beta tubulin are still almost impossible to see. We didn't bother trying to sort out the seam during the averaging process.

Example output

Example of easymode segment microtuble output overlaid on a tomogram from EMPIAR-11899 (FIB-milled D. discoideum), a dataset which was not used to train this model.