[3dem] ab initio reconstruction of multiple structures when each structure has a different point group symmetry

[Peter Doerschuk]

Dear Steve: Thank you very much for all this information!  I posted here because I have access to several software systems not just EMAN2.  But I had more success looking at the EMAN2 documentation so I used that as an example.  Best regards, Peter

________________________________
From: Ludtke, Steven J. <sludtke at bcm.edu>
Sent: Monday, August 31, 2026 2:59 PM
To: Peter Doerschuk <pd83 at cornell.edu>
Cc: 3dem at ncmir.ucsd.edu <3dem at ncmir.ucsd.edu>
Subject: Re: [3dem] ab initio reconstruction of multiple structures when each structure has a different point group symmetry

Hi Peter,
FYI - there's a separate mailing list (you'll find it at https://urldefense.com/v3/__https://groups.google.com/g/eman2__;!!Mih3wA!A8D45nYn7Yil1RZq5KNMezA51-3BSwaxoxhNLyw_I0ZQaPmzVSo0MFWxQsEZ1fEn_6i6EiI8n3B2aldJdQ$ ) for EMAN2 (https://urldefense.com/v3/__https://eman2.org__;!!Mih3wA!A8D45nYn7Yil1RZq5KNMezA51-3BSwaxoxhNLyw_I0ZQaPmzVSo0MFWxQsEZ1fEn_6i6EiI8n3BkIl9Itg$ <https://urldefense.com/v3/__https://eman2.org/__;!!Mih3wA!A8D45nYn7Yil1RZq5KNMezA51-3BSwaxoxhNLyw_I0ZQaPmzVSo0MFWxQsEZ1fEn_6i6EiI8n3A3-uBNOg$ >) specific questions. The mailing list has a searchable archive, so if there is anything you don't find in the wiki, search the mailing list before asking a new question.

Anyway, a typical approach would be to:
1) generate an initial model for each symmetry using the full set of particles (e2initialmodel_sgd.py) first. This need not be a good or high resolution map at all, just something rough with the correct symmetry.
2) run e2refinemulti.py with both initial models and the correct symmetry for each. That will simultaneously refine both maps, assigning each particle to the map it best matches in each iteration. Probably 4-5 iterations to get it to converge. Accuracy of particle classification will depend on how easy it is to distinguish between the two groups mathematically. In most cases there will be a certain amount of misclassification no matter what you do, unless you opted to just eliminate all of the particles in the ambiguous zone.

There are several other approaches you could use instead, such as running a single refinement with a single initial model and a high keep value and no symmetry, then after it's done, run e2refine_split.py or use e2gmm.py (deep learning heterogeneity analysis) to split the particles into each group. The advantage there would be that you get clear diagnostics about how clearly the two sets of particles break themselves into two groups.

Regardless of which approach you use, you should take the separated sets of particles and run refinements for a few iterations on each particle subset with the correct imposed symmetry independently to get the best structure for each group.

You could optionally follow that with e2evalparticles to try and identify any outliers WRT the final refined maps and get more homogeneous subsets.

In most cases in the end it comes down to whether the particles are mathematically well discriminated or not. If they are, then any classification method will probably work well. If they aren't then all methods will probably struggle to get accurate per-particle groups. However, that does NOT mean that you can't get accurate structures for each of the two groups. While this isn't maximum likelihood, you can think of it in those terms. If you have a continuous distribution with two overlapping Gaussian blobs, it is straightforward to find the location of the two peaks (the 2 structures) but particles between the peaks have a very high chance of individually being misclassified.

Some useful material (https://urldefense.com/v3/__https://eman2.org__;!!Mih3wA!A8D45nYn7Yil1RZq5KNMezA51-3BSwaxoxhNLyw_I0ZQaPmzVSo0MFWxQsEZ1fEn_6i6EiI8n3BkIl9Itg$ <https://urldefense.com/v3/__https://eman2.org/__;!!Mih3wA!A8D45nYn7Yil1RZq5KNMezA51-3BSwaxoxhNLyw_I0ZQaPmzVSo0MFWxQsEZ1fEn_6i6EiI8n3A3-uBNOg$ >):

Single particle tutorial:
https://urldefense.com/v3/__https://blake.bcm.edu/dl/EMAN2/EMAN2_SingleParticleTutorial.pdf__;!!Mih3wA!A8D45nYn7Yil1RZq5KNMezA51-3BSwaxoxhNLyw_I0ZQaPmzVSo0MFWxQsEZ1fEn_6i6EiI8n3A6Qy3TUw$ 

e2gmm GUI tutorial:
https://urldefense.com/v3/__https://blake.bcm.edu/emanwiki/doku.php?id=eman2:e2gmm__;!!Mih3wA!A8D45nYn7Yil1RZq5KNMezA51-3BSwaxoxhNLyw_I0ZQaPmzVSo0MFWxQsEZ1fEn_6i6EiI8n3Cyxcd-sA$ 

The mailing list:
https://urldefense.com/v3/__https://groups.google.com/g/eman2__;!!Mih3wA!A8D45nYn7Yil1RZq5KNMezA51-3BSwaxoxhNLyw_I0ZQaPmzVSo0MFWxQsEZ1fEn_6i6EiI8n3B2aldJdQ$ 


---
Steven Ludtke, Ph.D. <sludtke at bcm.edu>                      Baylor College of Medicine
Charles C. Bell Jr., Professor of Structural Biology        Dept. of Biochemistry
Deputy Director, Advanced Technology Cores                  and Molecular Pharmacology
Academic Director, CryoEM Core
Co-Director CIBR Center


On Aug 31, 2026, at 7:00 PM, Peter Doerschuk via 3dem <3dem at ncmir.ucsd.edu> wrote:

***CAUTION:*** EXTERNAL EMAIL
Only click links or open attachments you know are safe.
Report Phish<https://us-phishalarm-ewt.proofpoint.com/EWT/v1/GKvG1w!SH0yqwc8rEbQ_BFu5UoRsd6VZI9_owx0TdXS8oFcm4tRUm172XZQfcMkor9tQBR5gigSRIUWxDaOFbtUWbu_mUfmXePX-BoM494JcoAWQv0n0yAro47NJlJwFxxBzY28P1N6fIXSqLJJ9w$ >
ACTUAL SENDER ADDRESS: 3dem-bounces at ncmir.ucsd.edu<mailto:3dem-bounces at ncmir.ucsd.edu>
===============================================================
Dear 3DEM: For an engineering problem, I would like to do simultaneous ab initio reconstruction of two structures with different point group symmetries from a stack of unlabeled cryo EM images.  For instance, one structure has icosahedral symmetry, one structure has tetrahedral symmetry, and which images belong to which structure is not known.  I see e2refinemulti.html in EMAN2 but e2refinemulti.html is for refinement while I need ab initio model generation.  I would be very grateful for advice!  Thank you very much!  Best regards, Peter Doerschuk (pd83 at cornell.edu<mailto:pd83 at cornell.edu>)

_______________________________________________
3dem mailing list
3dem at ncmir.ucsd.edu<mailto:3dem at ncmir.ucsd.edu>
https://urldefense.com/v3/__https://mail.ncmir.ucsd.edu/mailman/listinfo/3dem__;!!GKvG1w!bACPRHT5EovIB5hqAaRd2XTlCduYDFcSteSEVHKZIjw9mYPEj2tHfLzkXbBIkVqGFJVvsiDcvDvO-zUboM9Z$

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.ncmir.ucsd.edu/pipermail/3dem/attachments/20260902/095eb429/attachment.html>