Do I understand correctly that for the homogeneous reconstruction only use
a) previous 3D refinement particles as input
b) using the same mask as we used for the previous 3D refinement
c) should get exactly the same resolution (as for previous 3D refinement) as an outcome of this protocol?
Because when I run the homogeneous reconstruction only protocol using 3D refinement particles as input the result resolution differs from the 3D refinement sufficiently.
So I suspect the problems in alignment particles information.
@Dmitry, may I suggest turning on the “Optimize FSC mask” parameter if you haven’t already; mask tightening worked for me in an issue similar to yours. Best,
Thank you - I have tried that - no changes.
Moreover, the result of the homogeneous reconstruction only protocol provides the same results with/without a mask (and with Optimize FSC mask parameter as well).
Homogeneous Reconstruction should replicate the results of standard 3D Refinement. However, it won’t replicate the results of non-uniform refinement if adaptive marginalization is enabled. This is because reconstruction only does not marginalize over alignments. Otherwise, the conditions you outlined in the first post are correct. Which refinement job are you comparing the reconstruction to?
As well, note that you must be careful about which mask is being used: the final refinement iteration shows two plots, one with the static/dynamic mask, and the other with the auto-tightened mask. The auto-tightened mask is output under the result name of mask_fsc_auto, and the static/dynamic mask used for the first FSC calculation is output under mask_fsc. Both of them are output so that the masks used are available for downstream work. To input the specific mask to the Reconstruction Only job, you must use low-level results which are documented in our guide.
It is an advanced parameter, so you must be viewing the job builder with advanced mode on. It’s titled “Adaptive Marginalization”, and is on by default but can be disabled.
I disabled the “Adaptive Marginalization” function, but the final reconstruction is still much worse for several proteins from different projects.
Do you know what else can cause this trouble?
Could you confirm that the homogenous reconstruction only works well after 3D non-uni alignment on CS version: v3.3.1+220118?
Thank you!
Kind regards,
Dmitry
p.s. I want to clarify that I am using the CS via SCIPION, and previously the homogeneous reconstruction worked (with perhaps may complicate the case)
p.p.s Tried with the CS separately - with the same result.
Apologies, I should clarify that there are other parameter choices that will affect whether the FSC results are exactly the same, “minimize over per-particle scale” being one. Could you send a screenshot of the inputs/parameters tabs of a refinement and associated reconstruction job where the resolution is significantly different? It may be easier to identify the culprit this way.
The issue was solved. The issue was connected to python.
Unfortunately, I can not provide more details than that.
At the same time, you may contact Yunior C. Fonseca Reyna (cfonseca@cnb.csic.es), who solved this issue inside the SCIPION shell.