14. Tomo refienment cycle(s)

Now that we saw the best resolution at the end of high resolution 3D refinement and post processing we reached is 5.08 A. In this section, we describe tomo refinement steps that include the optimisation of per-particle CTF parameters, and the modelling of beam-induced deformations to further improve the resolution. A tomo refinement cycle consists of two steps: CTF refinement, and Bayesian polishing. These steps go in tandem and their order of execution is irrelavent. Below, we describe one complete tomo refinement cycle, but in practice, one could carry out multiple cycles until no further improvment in resolution as well as the density is obtained. However, care should be taken not to over do this to avoid fitting to noise.

Note on Inputs and Outputs

Tomo refinement should be carried out using the subtomograms extracted at bin 1 level. That means the pixel size should be the original pixel size of the motion corrected images. The reference map and the alignment mask should also be reconstructed at this binning level.

As inputs, an optimisation_set.star, reference halfmap Reconstruct/jobxxx/half1.mrc, a file with FSC data postporcess.star, and an alignment mask at bin 1 align_mask.mrc are required. For the intial cycle, usually, an optimisation_set.star is not available. In that case, we can input tomograms.star and particles.star files, directly.

Each step (CTF refinement, or Bayesian polishing) in one tomo refinement cycle will output an optimisation_set.star. We can use that as input for the next step. We demonstrated how we did this below.

14.2. Reconstruct particle (relion.reconstructparticletomo)

Input params.	Bin 1
Input optimisation set:	`CtfRefine/job024/optimisation_set.star`
Input tomogram set:
Input particle set:
Box size (binned pix):	512
Cropped box size (binned pix):	192
Binning factor:	1
Symmetry:	`C6`
Outputs:
3D reference density map:	`Reconstruct/job025/merged.mrc`
half1 density map:	`Reconstruct/job025/half1.mrc`
half2 density map:	`Reconstruct/job025/half2.mrc`

14.3. Post process (relion.postprocess)

Input params.	Bin 1
One of the 2 unfiltered half-maps	`Reconstruct/job025/half1.mrc`
Solvent mask:	`fsc_mask.mrc`
Outputs:
Postprocessed starfile:	`Postprocess/job026/postprocess.star`
Postprocessed map:	`Postprocess/job026/postprocess.mrc`
Postprocessed masked-map:	`Postprocess/job026/postprocess_masked.mrc`

Gold-standard FSC showed the new resolution is 4.63 A. Next, we carried out the second step of the current tomo refinement cycle - Bayesian polishing using outputs from above jobs.

Bayesian polishing:

The input parameters we used are given in the images below.

At the end of the job, the output files optimisation_set.star, tomograms.star, particles.star and motion.star are written to Polish/job028 folder. The motion.star file contains motion trajectories of individual particles, particles.star contains the new particle positions, and tomograms.star file contains the updated tilt series alignment.

As before, we assessed the results by reconstructing a particle followed by postprocessing.

14.4. Reconstruct particle (relion.reconstructparticletomo)

Input params.	Bin 1
Input optimisation set:	`Polish/job028/optimisation_set.star`
Input tomogram set:
Input particle set:
Box size (binned pix):	512
Cropped box size (binned pix):	192
Binning factor:	1
Symmetry:	`C6`
Outputs:
3D reference density map:	`Reconstruct/job029/merged.mrc`
half1 density map:	`Reconstruct/job029/half1.mrc`
half2 density map:	`Reconstruct/job029/half2.mrc`

14.5. Post process (relion.postprocess)

Input params.	Bin 1
One of the 2 unfiltered half-maps	`Reconstruct/job029/half1.mrc`
Solvent mask:	`fsc_mask.mrc`
Outputs:
Postprocessed starfile:	`Postprocess/job030/postprocess.star`
Postprocessed map:	`Postprocess/job030/postprocess.mrc`
Postprocessed masked-map:	`Postprocess/job030/postprocess_masked.mrc`

We observed a clear improvement in resolution from 4.63 A to 4.25 A.

Once a tomo refinement cycle is complete, it is best practice to run a 3D refinement using the improved alignment (tomograms) and particles. For that, we extracted bin1 subtomograms using the Polish/job028/optimisation_set.star.

14.6. Extraction of pseudosubtomograms (relion.pseudosubtomo)

Input params.	Bin 1
Input optimisation set:	`Polish/job028/optimisation_set.star`
Box size (binned pix):	512
Cropped box size (binned pix):	192
Binning factor:	1
Outputs:
Pseudosubtomograms star file:	`Extract/job031/optimisation_set.star`

14.7. 3D refinement (relion.refine3d.tomo)

Input params.	Bin 1
Input optimisation set:	`Extract/job031/optimisation_set.star`
Reference map:	`Reconstruct/job029/half1.mrc`
Reference mask (optional):	`align_mask.mrc`
Ref. map is on absolute greyscale?	Yes
Resize reference if needed?	Yes
Initial low-pass filter (A):	4.5
Symmetry:	`C6`
Do CTF-correction?	Yes
Ignore CTFs until first peak?	No
Mask diameter (A):	230
Mask individual particles with zeros?	Yes
Use solvent-flattened FSCs?	Yes
Use Blush regularisation?	No
Initial angular sampling:	1.8
Initial offset range (pix):	5
Initial offset step (pix):	1
Local searches from auto-sampling:	1.8
Relax symmetry:
Use finer angular sampling faster?	No
Prior width on tilt angle (deg):	10
Use parallel disc I/O?	Yes
Number of pooled particles:	30
Skip padding?	No
Skip gridding?	Yes
Pre-read all particles into RAM?	No
Copy particles to scratch directory:
Combine iterations through disc?	No
Use GPU acceleration?	Yes
Number of MPI procs:	5
Number of threads:	8
Outputs:
Achieved Resolution (A)	4.39
Refined particles	`Refine3D/job032/run_data.star`
Optimisation set	`Refine3D/job032/run_optimisation_set.star`
Refined averaged density map:	`Refine3D/job032/run_class001.mrc`
half1 density map:	`Refine3D/job032/run_half1_class001_unfil.mrc`
half2 density map:	`Refine3D/job032/run_half2_class001_unfil.mrc`

At the end of the 3D refinement, we again ran Reconstruct particle followed by Postprocess to reach 4.25 A resolution map.

We refer you to the Relion documentation for more details.

Relion Tomo refinement documentation.