M reconstruction deteriorates after shifting species

[deadlift385]

Hi Alister,

I cannot get the same quality reconstruction after shifting a species.

Step 1) Start with an M refined reconstruction at 3.6 A resolution (box = 750, apix = 1.6).

Step 2a) Shift the species (MTools shift_species), which gives me half maps that look fine and are centered as intended. Step 2b) Calculate the FSC after cropping out a box (256) at the center of the shifted half maps and applying a spherical mask. The resolution is 3.9 A, as expected.

Step 3a) Run MCore without refinement. Step 3a) Calculate the FSC after cropping out a box (256) at the center of the shifted half maps and applying a spherical mask. The resolution is 6.7 A, much worse.

Step 1)

MCore \ --population m_run01/population_01.population \ --iter 0

2024-10-17 18:32:16.890 Global resolution is 3.597

Step 2a)

MTools shift_species \ --population m_run01/population_01.population \ --species m_run01/species/species_01_61d3e558/species_01.species \ --x 132 --y 298 --z 194

Step 2b)

Resolution from half maps after cropping and applying spherical mask = 3.9 (calculated with eman2 and relion)

Step 3a)

MCore \ --population m_run01/population_01.population \ --iter 0

Step 3b)

Resolution from half maps after cropping and applying spherical mask = 6.7 (calculated with eman2 and relion)

It's not obvious to me why one would get a substantially worse reconstruction (locally) after shifting the species and without refining anything. Am I missing something here? Is there some hidden weighing going on in the reconstruction step that gets messed up after shifting the species?

All the best, Simon

[alisterburt]

Hi @deadlift385 ! Thanks for the report - I'm not familiar with the move_species code myself so will have to take a look but I agree with your assessment, I don't see why it would manage to make a reconstruction after shifting the species then fail to reproduce it when running MCore...

In the meantime, you could try applying shifts to your input STAR file for species creation following the logic here

https://gist.github.com/alisterburt/5fa3e867f8004171fff57579b379ffa9

This script allows for applying multiple transformations at once, yielding n output particles per input particle. If you want to skip the re-orientation supply an identity matrix or comment out the code 🙂

[deadlift385]

Thank you for the link to your script! I can make Relion star files after using MTools expand_symmetry and MTools shift_species, this works all fine. After symmetry expansion M gives me the same reconstruction as compared to imposing symmetry (expected). Getting the same quality reconstructions, and exported subtomograms for classification, at shifted positions is the issue. I can do a new species creation, but I would have to see how stable the M refinement of those would be.

[alisterburt]

@deadlift385 right, I'm suggesting that you remove the use of M integrated tools, apply the shifts yourself then create a new species in M and see if the problems persist - this would allow us to say whether the problem is with shift_species or something strange with your data (unlikely) 🙂

[deadlift385]

In fact, I have already done this, created a new species with the new positions and angles, then run M without refinement (--iter 0). Same resolution as after shifting the original species and running M, 6.4 instead of 3.9. And the maps look correspondingly worse.

[alisterburt]

@deadlift385 hmm, could you share a few xy slices of the maps? It seems like the resolution must be worse where your new map is centered... the only missing control would be to check the FSC of your newly centered region in the original map