FilamentTools

David Li, while he was an Mphil student in our group, wrote the FilamentTools script that performs bi-hierarchical classification on filaments and 2D class assignments that is described in our paper on time-resolved cryo-EM of tau filament formation.

You can run the FilamentTools from the Subset selection jobtype on the relion-5 GUI (set option to Yes on Filament tab). The program will take a little while to run. If you have millions of particles, it may fail. If that’s the case, please try and divide the data set into multiple smaller subsets.

Once the calculation has finished, display the resulting logfile.pdf with the bi-hierachical classification. Horizontal separation is based on individual filaments in the micrographs (from picked start-end coordinates). Vertical separation is based on the 2D classification. Display from the GUI pull-down button also the output file run_optimiser.star. On the pop-up display menu, from the pull-down menu, select rlnClassNumber, but don’t tick the box to sort on this number; instead just tick the box to display this label on the images. Then you will see the classes in the same order as (from top to bottom) on the right-hand side of the PDF. Perfect 2D classifications would have clean blocks of 2D classes (with different viewing angles of the same filament types) belong to a clean subset of the filaments: resulting in rectangular yellow blocks in the histogram. In practice, 2D classification is not perfect, but one can often still see the rectangular boxes. Assuming that filament types do not change within individual filaments, selecting subsets of the corresponding filaments from such boxes (based on the horizontal dendrogram at the top of the PDF) will then lead to cleaner classes than merely selecting 2D classes. For example, you can clearly see 4 blocks in Supplementary Figure 5 of the tau paper. A fifth one is also visible, but may have been pulled out in a second round.

Once you have the PDF, you can continue the same job on the GUI, but use a positive number for “minimum class size” on the Select GUI, to write out multiple star files that correspond to subsets of the dendrogram on filament type (vertical separation, at the top of the PDF). You can specify lower thresholds (from the GUI; you can run multiple Continue jobs until you find the correct threshold) to get more subset star files on the GUI. You may then need to combine some of the star files (using relion_star_handler), based on the threshold used and the shape of the dendrogram. The star files are numbered from the left to the right of the dendrogram. Sometimes it’s a bit tricky to see which one is which, but looking at the size of the star files, and possibly also looking inside the STAR files to see which rlnClassNumber the segments belong to, will help.

Particles selected can be used again for a second round of 2D classification and use of the same FilamentTools, if the results from the first run did not yet yield homogeneous classes.