ERROR IN PROTEIN-PROTEIN STRUCTURE PREDICTION USING ALPHA-FOLD

[daniel.sussman@…]

Dear Sirs,
When running ALPHA-FOLD for a protein-protein structure prediction in ChimeraX, I got the following error message (see enclosed pdf). I would very much appreciate if you could tell me the error source and a possible solution.

I'am also enclosing the fasta file which I used in this run.
Thanks in advance,
Daniel K. Sussman-Villaverde

[Tom Goddard]

The error was probably a temporary problem with connecting to the conda repository to get OpenMM or maybe a temporary networking problem with Google Colab. I just ran the prediction with your sequences from ChimeraX 1.8 and it worked fine, took about 20 minutes.

In the future be sure to copy and paste the exact error message when reporting the problem. Without the exact error there is almost nothing we can do to diagnose the problem. Also use ChimeraX menu entry Help / Report a Bug so we know what operating system and version of ChimeraX you are using.

[daniel.sussman@…]

Dear Dr. Goddard,
We tried several times to run this calculation, not just once.
It run well when we did not ask for an energy optimization.
What could make the difference?
Thanks again,
Daniel K. Sussman
________________________________
De: ChimeraX <ChimeraX-bugs-admin@cgl.ucsf.edu>
Enviado: mércores, 30 de outubro de 2024 19:16
Para: Sussman Villaverde Daniel <daniel.sussman@rai.usc.es>; goddard@cgl.ucsf.edu <goddard@cgl.ucsf.edu>
Asunto: Re: [ChimeraX] #16218: ERROR IN PROTEIN-PROTEIN STRUCTURE PREDICTION USING ALPHA-FOLD

#16218: ERROR IN PROTEIN-PROTEIN STRUCTURE PREDICTION USING ALPHA-FOLD
-------------------------------------------+-------------------------
          Reporter:  daniel.sussman@…      |      Owner:  Tom Goddard
              Type:  defect                |     Status:  closed
          Priority:  moderate              |  Milestone:
         Component:  Structure Prediction  |    Version:
        Resolution:  worksforme            |   Keywords:
        Blocked By:                        |   Blocking:
Notify when closed:                        |   Platform:  all
           Project:  ChimeraX              |
-------------------------------------------+-------------------------
Changes (by Tom Goddard):

 * resolution:   => worksforme
 * status:  assigned => closed

Comment:

 The error was probably a temporary problem with connecting to the conda
 repository to get OpenMM or maybe a temporary networking problem with
 Google Colab.  I just ran the prediction with your sequences from ChimeraX
 1.8 and it worked fine, took about 20 minutes.

 In the future be sure to copy and paste the exact error message when
 reporting the problem.  Without the exact error there is almost nothing we
 can do to diagnose the problem.  Also use ChimeraX menu entry Help /
 Report a Bug so we know what operating system and version of ChimeraX you
 are using.
--
Ticket URL: <https://www.rbvi.ucsf.edu/trac/ChimeraX/ticket/16218#comment:1>
ChimeraX <https://www.rbvi.ucsf.edu/chimerax/>
ChimeraX Issue Tracker

[Tom Goddard]

Ah, I was not thinking. ChimeraX does not do energy minimization by default when running AlphaFold predictions, and the test I ran did not enable it. Only energy minimization uses OpenMM. So let me try again the run with energy minimization.

The reason ChimeraX does not do energy minimization with AlphaFold predictions by default is that it is both unreliable (often OpenMM dies during the minimization making the prediction fail saying it could not converge after 100 steps due to clashes in the AlphaFold model), and also it has little effect at improving the models.

[Tom Goddard]

I just ran AlphaFold prediction on your trimer with energy minimization in ChimeraX 1.8 and OpenMM installed correctly. I've attached the install_log.txt file. As I mentioned before there is no way to help you if you don't provide the actual error. You can download the install_log.txt by clicking on the folder icon on the left edge of the AlphaFold Run panel and clicking the "..." when you hover over the install_log.txt and selecting Download. It will go into your ~/Downloads/ChimeraX/prediction_N folder. Then you can mail it in reply to this bug report email to attach it.

My run with energy minimization (on Nvidia T4) took a very long time to do the energy minimization, about an extra hour to do the minimization beyond the 20 minutes for the install and prediction. So it made the run take 4 times longer, and it just moved some side chains from clashing by doing the minimization. So I don't think it makes sense do the energy minimization -- it is poorly implemented by AlphaFold and OpenMM.

[Tom Goddard]

Install log from my test run that shows OpenMM installs without errors.