10. Initial 3D Refinement

Now that we have bin6 subtomograms and a 3D reference (from Reconstruct particle) to start the initial 3D refinement, which uses the RELION’s auto-refine engine. In general, the 3D refinement should be carried out from particles with hight binning factors to low binning factors. This strategy will not only improve the processing time, but also help to reach a better convergence (ideally, global minimum). Another advantage is that, the stepwise processing will help you to decide the resolution limit of your data sometimes before reaching the bin 1 level.

For the tutorial data, we have carried out 3D refinement at two bining levels. Our initial 3D refinement at bin6 level reached the Nyquist resolution, which is 16.2 A. The SNR column in the Refine3D/job009/run_model.star showed that the signal clearly goes beyond the Nyquist. At that point, we decided the carry out another 3D refinement with bin 2 particles. The bin 2 particles were extracted using the refined particle-coordinates (from Refine3D/job009/run_data.star), and a new reference was generated with Reconstruct particle job. All the input parameters we used are summarised in the tables below for your easy reference. We refer you to the RELION documentation for details of those parameters.

Relion Initial 3D refinement documentation.

10.1. Extraction of pseudosubtomograms

Input params.	Bin 6	Bin 2
Input optimisation set:		`Refine3D/job009/run_optimisation_set.star`
Input tomogram set:	`Tomograms/job006/tomograms.star`
Input particle set:	`particles.star`
Box size (binned pix):	192	256
Cropped box size (binned pix):	96	128
Binning factor:	6	2
Outputs:
Pseudosubtomograms star file:	`Extract/job008/optimisation_set.star`	`Extract/job010/optimisation_set.star`

10.2. Reconstruct Particle

Input params.	Bin 6	Bin 2
Input optimisation set:		`Extract/job010/optimisation_set.star`
Input tomogram set:	`Tomograms/job006/tomograms.star`
Input particle set:	`particles.star`
Box size (binned pix):	192	256
Cropped box size (binned pix):	96	128
Binning factor:	6	2
Symmetry:	`C6`	`C6`
Outputs:
3D reference density map:	`Reconstruct/job007/merged.mrc`	`Reconstruct/job011/merged.mrc`
half1 density map:		`Reconstruct/job011/half1.mrc`
half2 density map:		`Reconstruct/job011/half2.mrc`

10.3. 3D refinement (auto-refine)

Input params.	Bin 6	Bin 2
Input optimisation set:	`Extract/job008/optimisation_set.star`	`Extract/job010/optimisation_set.star`
Reference map:	`Reconstruct/job007/merged.mrc`	`Reconstruct/job011/half1.mrc`
Reference mask (optional):
Ref. map is on absolute greyscale?	Yes	Yes
Resize reference if needed?	Yes	Yes
Initial low-pass filter (A):	60	20
Symmetry:	`C6`	`C6`
Do CTF-correction?	Yes	Yes
Ignore CTFs until first peak?	No	No
Mask diameter (A):	500	230
Mask individual particles with zeros?	Yes	Yes
Use solvent-flattened FSCs?	No	No
Use Blush regularisation?	No	No
Initial angular sampling:	7.5	7.5
Initial offset range (pix):	5	5
Initial offset step (pix):	1	1
Local searches from auto-sampling:	1.8	1.8
Relax symmetry:
Use finer angular sampling faster?	No	No
Prior width on tilt angle (deg):	10	10
Use parallel disc I/O?	Yes	Yes
Number of pooled particles:	30	30
Skip padding?	No	No
Skip gridding?	Yes	Yes
Pre-read all particles into RAM?	No	No
Copy particles to scratch directory:
Combine iterations through disc?	No	No
Use GPU acceleration?	Yes	Yes
Number of MPI procs:	5	5
Number of threads:	6	6
Outputs:
Achieved Resolution (A)	16.54	5.49
Refined particles	`Refine3D/job009/run_data.star`	`Refine3D/job012/run_data.star`
Optimisation set	`Refine3D/job009/run_optimisation_set.star`	`Refine3D/job012/run_optimisation_set.star`
Refined averaged density map:	`Refine3D/job009/run_class001.mrc`	`Refine3D/job012/run_class001.mrc`
half1 density map:	`Refine3D/job009/run_half1_class001_unfil.mrc`	`Refine3D/job012/run_half1_class001_unfil.mrc`
half2 density map:	`Refine3D/job009/run_half2_class001_unfil.mrc`	`Refine3D/job012/run_half2_class001_unfil.mrc`

The bin 2 refinement converged to 5.49 A resolution and is also the Nyquist resolution. The density of the refined map looked to be 5 - 6 A, and the side chain features were starting to emerge. Below we include some figures of the Refine3D/job012/run_class001.mrc for your reference.

Example image — Orthogonal slices of the density