Once you are satisfied with your particle picks after following the previous steps, you can import them into RELION-5 for subtomogram averaging. For a detailed, step-by-step tutorial on subtomogram averaging with RELION-5, we refer you to the official tutorial as well as the TomoGuide. Here, we will focus on the steps that are specifically related to MemBrain-pick, namely, the importing of particle coordinates as well as how to run constrained angular searches during 3D refinement and classification.
After the prediction step, MemBrain-pick should have generated one STAR file per segmented membrane, similar to the one below:
data_
loop_
_rlnCoordinateX #1
_rlnCoordinateY #2
_rlnCoordinateZ #3
_rlnAngleRot #4
_rlnAngleTilt #5
_rlnAnglePsi #6
5904.217285 4464.462891 2593.735107 87.142778 91.651887 30.010083
5907.051758 4423.755371 2330.111572 76.163535 97.265316 27.178633
5946.418457 4455.405762 2078.243164 84.601583 92.932160 28.427001
In order to import into RELION-5, you should use your favorite text editor or a little script to generate a "master" STAR file associating all the per-membrane STAR files (as exemplified above) to their respective tomograms, which should look like the example below. Let's call this file particles_membrain.star:
data_particles
loop_
_rlnTomoName #1
_rlnTomoImportParticleFile #2
tomo_0001 membrain-pick/membrain_pick_output/tomo_0001_T1S1M01_clusters.star
tomo_0002 membrain-pick/membrain_pick_output/tomo_0002_T2S1M01_clusters.star
tomo_0002 membrain-pick/membrain_pick_output/tomo_0002_T2S1M02_clusters.star
tomo_0003 membrain-pick/membrain_pick_output/tomo_0001_T3S1M01_clusters.star
NOTE: The paths indicated in the rlnTomoImportParticleFile column should be relative to your RELION project directory.
You now need to consider what is your unbinned pixel size, so that the coordinates are scaled properly. Internally, MemBrain-pick assumes the rlnCoordinate<X/Y/Z> columns are in non-centered Å; RELION-5, however, expects these values to be in unbinned pixels if the columns have this name (not to be confused with the default rlnCenteredCoordinate<X/Y/Z>Angst columns). It is therefore just a matter of applying a scaling factor, which should be supplied in the Coordinates tab of the RELION Import job. If your unbinned pixel size is 1.96 Å, you should therefore supply 1/1.96 = 0.51020408 as your scaling factor to convert the coordinates back to pixels.
Assuming you already have an existing RELION tomography project with all the proper tilt series data imported/processed, you can then import the particles by running an Import job as follows:
Currently, the membrane labels (e.g. T1S1M01) are lost when importing into RELION. Preserving this label is on our TO-DO list. Optionally, you can write a little script to insert a label into each of your particle STAR files, which RELION will preserve upon importing.
Once coordinates and initial Euler angles have been imported, it is highly recommend to run a Reconstruct particle job to see if RELION properly understood the particle STAR files. Since the in-plane angles are randomized at this point, you should obtain an average that is (roughly) rotationally averaged around the Z axis. For details of running this job, please refer to the official tutorial or the TomoGuide. It is also recommended to visually inspect the imported particle coordinates and orientations in ArtiaX as explained in the TomoGuide.
Particle picks from MemBrain-pick are already pre-oriented normal to the membrane, but are randomly oriented within the membrane plane. This means that you don't need to run a full global search when running alignment and classication jobs in RELION. You can just do a full in-plane search, and a tiny local cone-like search around the vector normal to the membrane.
For doing this, in the Sampling (3D classification) or Auto-sampling (3D auto-refine) tab you first define the global angular ampling interval. In the example below, this is 7.5 degrees:
Then in the Running tab you should supply additional arguments to impose a small Gaussian prior on the rlnAngleRot and rlnAnglePsi Euler angles. In the example below, we impose a prior with a 3-degree standard deviation (--sigma_tilt 3 --sigma_psi 3):
Along the refinement and classification rounds, you should always inspect your particle orientations in ArtiaX as explained in the TomoGuide to make sure the angular searches are behaving as expected.
As is usual, once your subtomogram average improves and angles start to converge, you can run 3D auto-refine jobs with only local searches. For this, you just need to supply the same initial sampling interval for both global and local searches, which is 3.7 degrees in the example below. There is no need to provide additional arguments in the Running tab now.
