Tomo frame alignment

relion has also implemented the analogous to Bayesian polishing in 2D for tomography. This procedure refines the projections that map 3D space onto the images of the tilt series. Optionally, the beam-induced motion trajectories of the particles and deformations can also be estimated. For a complete description of the arguments, check relion_tomo_align program.

If you are running this job without following the tutorial, please, note the requirements for the reference map as described for Tomo CTF refinement job in subsection Reference map at bin 1 and FSC data.

Running the job

The optimisation set file we are using as input has updated the reference mask file field in the previous Tomo CTF refinement job, so we don’t need to override any other input file.

On the I/O tab of the Tomo frame alignment job-type set:

Input optimisation set:


On the Polish tab, set:

Box size for estimation (pix)


Max position error (pix)


Align by shift only?


Alignment model

(Does not apply)

On the Motion tab, set:

Fit per-particle motion?


Sigma for velocity (Å/dose)


Sigma for divergence (Å)


Use Gaussian decay


On the Deformations tab set:

Estimate 2D?


This dataset has shown better results when deformations have not been estimated.

On the Running tab, set:

Number of MPI procs:


Number of threads:


Note that the per-particle motion estimation increases significantly the processing time. In our system it took around 2 hours.

Analysing the results

In the output folder FrameAlignTomo/job017 you will find new and files including the corrected tilt series alignment and particle positions and a trajectory set file with particle trajectories. Again, to assess the result, it is recommended to run a new Tomo reconstruct particle job, with FSC estimation, using the new parameters. Compared to the previous FSC estimation, we should observe a clear improvement and a resolution around 3.5Å.

Tomo refinement cycle

After running both tomo specific refinement steps, it is still recommended to run a new 3D auto-refine job to take advantage of the improved tomograms and particles. To that end, we need to construct a new set of pseudo-subtomos and reference maps as described in subsection Pseudo-subtomograms and reference map at bin 1. For the new 3D auto-refine job, same parameter as in subsection Running the auto-refine job at bin 1 apply except for:

On the Reference tab, set:

Initial low-pass filter (A)


On the Auto-sampling tab set:

Initial angular sampling:

0.9 degrees

This new 3D refinement spent 1 day in our system (4 GPU cards) and it should report a resolution around 3.4Å, completing the first tomo refinement cycle. If a new cycle of Tomo CTF refinement, tomo frame alignment and 3D auto-refine is performed, the user should reach around 3.3Å and finally converge to 3.2Å in the third cycle.