• Most programs are available in EMAN2 builds after 2026-06, but newer versions are typically better.
• The tutorial is only tested on Linux with Nvidia GPU and CUDA.
• For reference of the method, see the preprint.
• A separate tutorial about the same method applied to CryoET data can be found here.

In this tutorial, we use a refinement of TRPV1 from a CryoSPARC tutorial. In theory you can also do everything in EMAN2 from scratch, but this makes a clearer comparison with other software.

Start from the consensus refinement from CryoSPARC, and import the particles with:

e2convertrelion.py J35_008_particles.star --onestack particles/all_particles.hdf

Depending on the masking, the resolution should be around 2.7 to 3 Å. After the import finishes, you should find relavent files in the `r3d_00` folder.

First build the GMM representation from the 3D volume.

e2gmm_guess_n.py r3d_00/threed_00.hdf --evenodd --nopdb --thr 4.5 --maxres 3

Here the `–evenodd` option will build GMM for the two half-set separately. Pick a threshold for `–thr` so all protein density of interest are visible. `–nopdb` force it to write points as txt instead of pdb, since the pdb format limits the maximum number of atoms. This should produce `r3d_00/threed_00_gmm_even/odd.txt`.

Run the global refinement next.

e2gmm_refine_iter.py r3d_00/threed_00.hdf --startres 2.8 --initpts r3d_00/threed_00_gmm.txt --maskpp r3d_00/mask.hdf --jax --niter 3 --parallel thread:32 --sym c4

It is ok to use `r3d_00/threed_00_gmm.txt` even though the file does not exist, since the program will look for the even/odd versions automatically. Make sure to add `–jax` to call the JAX backend, as we are retiring the Tensorflow backend gradually. Without `–maskpp`, the program will use auto-generated masks. Here we keep the same masking for better comparison.

After the refinement, the global resolution should reach Nyquist, at 2.5 Å, according to the “gold-standard” FSC.