[3dem] Looking for feedback on deep learning-based post-processing

Mon Apr 13 18:25:18 PDT 2026

Great and useful discussion so far.
Re Alkin’s response. Thanks for the distinction between “pre-trained” methods, and methods that only train on the particle stack in question. There is “preexisting knowledge” in the latter case, and is as you say, "The statistical model, formulated through a set of neural network equations…”. These equations constitute the “experience” of the method.
I think it is fairly simple. Data is data. Images and maps that are deposited (especially if they could potentially be used to train new AI networks in a feedback loop) should be data-and-mathematical-algorithm-based (raw or sharpened) such that they can be reproduced. The more the better! AI-generated maps are an interpretation, just like we would make, based on “experience”. They can be just as useful, or as misleading, as human-brained interpretations. This is also true of maps derived from latent space analysis. Properly attributed AI maps used in figures are fine, especially side-by-side with the data-derived maps. There can then be a discussion of the reasons why a feature appears in an AI map that does not appear in the data-derived map, and vice versa.
Mike

From: 3dem <3dem-bounces at ncmir.ucsd.edu> on behalf of Paolo Swuec via 3dem <3dem at ncmir.ucsd.edu>
Date: Monday, April 13, 2026 at 4:29 AM
To: Collaborative Computational Project in Electron cryo-Microscopy <ccpem at jiscmail.ac.uk>, '3dem at ncmir.ucsd.edu' <3dem at ncmir.ucsd.edu>
Subject: [3dem] Looking for feedback on deep learning-based post-processing

BEWARE: This email originated outside of our organization. DO NOT CLICK links or attachments unless you recognize the sender and know the content is safe.

Dear CCPEMers and 3DEMers,

I hope this message finds you well. I am reaching out to seek your advice and perspective on a question that has been increasingly relevant in my work as a scientist as well as a reviewer. Specifically, I would greatly appreciate your feedback on how best to incorporate maps obtained through deep learning-based post-processing methods in SPA workflows.

While these approaches often produce visually improved maps (enhancing features such as connectivity and apparent resolution) I am particularly interested in understanding or even get your 2 cents on:

  *
Does anybody have data\maps showing clear signs of AI-hallucinations?
  *
If a paper includes map from deep learning-based post-processing method, must this be deposited on EMDB?
  *
What could be a reasonable validation tool to prove existence of new regions\loops derived from such maps and not seen in regularly post-processed maps? (We are lucky enough to have structural proteomics here, but what to do when this is not available?)

I am especially keen to hear about both positive experiences and any concerns or limitations you may have encountered in practice. I am NO expert at all in DL methods 😊 so please don’t shoot!

Thank you very much in advance for your time and insights. I believe this is an important topic for ensuring consistent and robust use of emerging computational methods within the field.

Best!

Paolo

Paolo Swuec
Head of National Facility for Structural Biology
Human Technopole
Palazzo Italia
Viale Rita Levi‑Montalcini, 1
20157 Milan, Italy
T. +39 02-30247442<tel:02-30247442>
M. +39 3427758814<tel:3427758814>
E. paolo.swuec at fht.org<mailto:paolo.swuec at fht.org>
humantechnopole.it<https://urldefense.com/v3/__http://www.humantechnopole.it/__;!!Mih3wA!GJToexaY1mi0S2ap0A98Hdi_WMPNEFmbI-nMZERrtS3gqlB70pbvhRVmKaZNI4pplffBIS61fiMMw7kd2SS2FpA$>
[cid:image861547.png at 7D3E7A85.6A4901DA]<https://urldefense.com/v3/__https://www.facebook.com/htechnopole__;!!Mih3wA!GJToexaY1mi0S2ap0A98Hdi_WMPNEFmbI-nMZERrtS3gqlB70pbvhRVmKaZNI4pplffBIS61fiMMw7kdAuIT6Lo$>
[cid:image313735.png at B8AD428C.DDEAA6E1]<https://urldefense.com/v3/__https://twitter.com/humantechnopole__;!!Mih3wA!GJToexaY1mi0S2ap0A98Hdi_WMPNEFmbI-nMZERrtS3gqlB70pbvhRVmKaZNI4pplffBIS61fiMMw7kdMLJNzSI$>
[cid:image897435.png at A7487D41.AB0BD1CB]<https://urldefense.com/v3/__https://www.linkedin.com/company/htechnopole__;!!Mih3wA!GJToexaY1mi0S2ap0A98Hdi_WMPNEFmbI-nMZERrtS3gqlB70pbvhRVmKaZNI4pplffBIS61fiMMw7kdhpubwVA$>
[cid:image285495.png at E4579D9F.10CE4F4D]<https://urldefense.com/v3/__https://www.instagram.com/humantechnopole/__;!!Mih3wA!GJToexaY1mi0S2ap0A98Hdi_WMPNEFmbI-nMZERrtS3gqlB70pbvhRVmKaZNI4pplffBIS61fiMMw7kdYtt4OSA$>
[cid:image259763.jpg at 587A226F.082AAE68]
Prima di stampare questa e-mail, pensa all’ambiente.
Before printing this e-mail, think of the environment
Legal Disclaimer
The information contained in or accompanying this email and any attachment thereto, is intended solely for the use of the stated recipient(s) and may contain information that is confidential and/or privileged. Any dissemination, distribution or copying of this email and any attachment by anyone who is not the stated recipient is strictly prohibited. If you receive this message in error, please notify the sender immediately and delete the message and any attachment from your system without retaining a copy.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.ncmir.ucsd.edu/pipermail/3dem/attachments/20260414/8d11ea1e/attachment-0001.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: image861547.png
Type: image/png
Size: 701 bytes
Desc: image861547.png
URL: <http://mail.ncmir.ucsd.edu/pipermail/3dem/attachments/20260414/8d11ea1e/attachment-0004.png>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: image313735.png
Type: image/png
Size: 680 bytes
Desc: image313735.png
URL: <http://mail.ncmir.ucsd.edu/pipermail/3dem/attachments/20260414/8d11ea1e/attachment-0005.png>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: image897435.png
Type: image/png
Size: 464 bytes
Desc: image897435.png
URL: <http://mail.ncmir.ucsd.edu/pipermail/3dem/attachments/20260414/8d11ea1e/attachment-0006.png>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: image285495.png
Type: image/png
Size: 721 bytes
Desc: image285495.png
URL: <http://mail.ncmir.ucsd.edu/pipermail/3dem/attachments/20260414/8d11ea1e/attachment-0007.png>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: image259763.jpg
Type: image/jpeg
Size: 64981 bytes
Desc: image259763.jpg
URL: <http://mail.ncmir.ucsd.edu/pipermail/3dem/attachments/20260414/8d11ea1e/attachment-0001.jpg>