Failing to complete dimer prediction
Dear all, I am trying to use ChimeraX to predict the structure of a dimer as part of my Master’s thesis. I was able to produce a model of the monomeric structure without issue have not been able to do the same for the dimer. I am currently subscribed to Colab Pro so had assumed that the 15.0GB GPU RAM would be sufficient. The structure is 1361 amino acids long but AlphaFold2 is still able to produce the first two models successfully. However, I have not been able to produce any iterations past this stage. Any guidance on how to complete the model would be much appreciated! Alternatively, I would also be glad to extract the output of the first two models (e.g. .pdb files, plDDT plots, PAE plots, MSA heatmaps) but because the prediction does not run to completion I don’t know how to access/extract these. I have really enjoyed using the program and hope to include the structure of the dimer in my thesis before I submit in three weeks’ time. Many thanks, Tal
Hi Tal, Please see these earlier posts about what to try when the calculation is large: <https://www.rbvi.ucsf.edu/pipermail/chimerax-users/2023-April/005119.html> <https://www.rbvi.ucsf.edu/pipermail/chimerax-users/2023-April/005120.html> Otherwise, I don't have any suggestions except to say that you can specify a results directory or folder. If any of the intermediate results are written to files (and I do not know if they are), they would appear in this directory. If you are using a recent version of ChimeraX, the AlphaFold tool will have an Options button, and clicking it will show further settings including "Results directory." For details see the help: <https://rbvi.ucsf.edu/chimerax/docs/user/tools/alphafold.html> <https://rbvi.ucsf.edu/chimerax/docs/user/tools/alphafold.html#predict> Best, Elaine ----- Elaine C. Meng, Ph.D. UCSF Chimera(X) team Department of Pharmaceutical Chemistry University of California, San Francisco
On May 1, 2023, at 6:38 AM, Tal Jeffrey via ChimeraX-users <chimerax-users@cgl.ucsf.edu> wrote:
Dear all,
I am trying to use ChimeraX to predict the structure of a dimer as part of my Master’s thesis. I was able to produce a model of the monomeric structure without issue have not been able to do the same for the dimer. I am currently subscribed to Colab Pro so had assumed that the 15.0GB GPU RAM would be sufficient. The structure is 1361 amino acids long but AlphaFold2 is still able to produce the first two models successfully. However, I have not been able to produce any iterations past this stage. Any guidance on how to complete the model would be much appreciated! Alternatively, I would also be glad to extract the output of the first two models (e.g. .pdb files, plDDT plots, PAE plots, MSA heatmaps) but because the prediction does not run to completion I don’t know how to access/extract these.
I have really enjoyed using the program and hope to include the structure of the dimer in my thesis before I submit in three weeks’ time.
Many thanks,
Tal
_______________________________________________ ChimeraX-users mailing list ChimeraX-users@cgl.ucsf.edu Manage subscription: https://www.rbvi.ucsf.edu/mailman/listinfo/chimerax-users
Hi Elaine, Thank you for your response. I kept trying the prediction and fortunately it has worked! If you don’t mind, may I ask you a few more questions to make sure I am interpreting the confidence metrics correctly? * To the best of my knowledge, no experimental data yet exists for the dimer I have modelled. In situations like these, does AlphaFold2 use homology modelling to generate a predicted structure based off of homologous templates? * In cases where proteins are predicted without experimental data, are the sequence coverage plots expected to have low sequence identity to query values? How would this change your interpretation of the prediction if the plDDT value was high (~80)? * Could you explain how the pTM score relates to the PAE values? * Could you explain how you would interpret the PAE plots for predictions of these two different proteins: [cid:image001.png@01D97C5E.305558D0] [cid:image002.png@01D97C61.05E2C580] Thank you again and apologies for the bombardment of questions! Tal From: Elaine Meng <meng@cgl.ucsf.edu> Date: Monday, 1 May 2023 at 16:22 To: Tal Jeffrey <tal.jeffrey@queens.ox.ac.uk> Cc: chimerax-users@cgl.ucsf.edu <chimerax-users@cgl.ucsf.edu> Subject: Re: [chimerax-users] Failing to complete dimer prediction Hi Tal, Please see these earlier posts about what to try when the calculation is large: <https://www.rbvi.ucsf.edu/pipermail/chimerax-users/2023-April/005119.html> <https://www.rbvi.ucsf.edu/pipermail/chimerax-users/2023-April/005120.html> Otherwise, I don't have any suggestions except to say that you can specify a results directory or folder. If any of the intermediate results are written to files (and I do not know if they are), they would appear in this directory. If you are using a recent version of ChimeraX, the AlphaFold tool will have an Options button, and clicking it will show further settings including "Results directory." For details see the help: <https://rbvi.ucsf.edu/chimerax/docs/user/tools/alphafold.html> <https://rbvi.ucsf.edu/chimerax/docs/user/tools/alphafold.html#predict> Best, Elaine ----- Elaine C. Meng, Ph.D. UCSF Chimera(X) team Department of Pharmaceutical Chemistry University of California, San Francisco
On May 1, 2023, at 6:38 AM, Tal Jeffrey via ChimeraX-users <chimerax-users@cgl.ucsf.edu> wrote:
Dear all,
I am trying to use ChimeraX to predict the structure of a dimer as part of my Master’s thesis. I was able to produce a model of the monomeric structure without issue have not been able to do the same for the dimer. I am currently subscribed to Colab Pro so had assumed that the 15.0GB GPU RAM would be sufficient. The structure is 1361 amino acids long but AlphaFold2 is still able to produce the first two models successfully. However, I have not been able to produce any iterations past this stage. Any guidance on how to complete the model would be much appreciated! Alternatively, I would also be glad to extract the output of the first two models (e.g. .pdb files, plDDT plots, PAE plots, MSA heatmaps) but because the prediction does not run to completion I don’t know how to access/extract these.
I have really enjoyed using the program and hope to include the structure of the dimer in my thesis before I submit in three weeks’ time.
Many thanks,
Tal
_______________________________________________ ChimeraX-users mailing list ChimeraX-users@cgl.ucsf.edu Manage subscription: https://www.rbvi.ucsf.edu/mailman/listinfo/chimerax-users
Hi Tal, We are not the AlphaFold developers, and I in particular do not have that much experience using it. Did you read the previous posts and try the web server (non ChimeraX, it was a separate website) or any of the other suggestions they contained? If the ChimeraX documentation does not have enough detail to clarify your questions, all I can suggest is to read the AlphaFold papers and information at the AlphaFold Database website (and links therein) to find the answers. There are links to several AlphaFold papers as well as the database in the ChimeraX AlphaFold documentation: <https://rbvi.ucsf.edu/chimerax/docs/user/tools/alphafold.html> Best, Elaine ----- Elaine C. Meng, Ph.D. UCSF Chimera(X) team Department of Pharmaceutical Chemistry University of California, San Francisco
On May 1, 2023, at 11:13 AM, Tal Jeffrey via ChimeraX-users <chimerax-users@cgl.ucsf.edu> wrote:
Hi Elaine,
Thank you for your response. I kept trying the prediction and fortunately it has worked!
If you don’t mind, may I ask you a few more questions to make sure I am interpreting the confidence metrics correctly? • To the best of my knowledge, no experimental data yet exists for the dimer I have modelled. In situations like these, does AlphaFold2 use homology modelling to generate a predicted structure based off of homologous templates? • In cases where proteins are predicted without experimental data, are the sequence coverage plots expected to have low sequence identity to query values? How would this change your interpretation of the prediction if the plDDT value was high (~80)? • Could you explain how the pTM score relates to the PAE values? • Could you explain how you would interpret the PAE plots for predictions of these two different proteins: <image001.png>
<image002.png>
Thank you again and apologies for the bombardment of questions!
Tal
From: Elaine Meng <meng@cgl.ucsf.edu> Date: Monday, 1 May 2023 at 16:22 To: Tal Jeffrey <tal.jeffrey@queens.ox.ac.uk> Cc: chimerax-users@cgl.ucsf.edu <chimerax-users@cgl.ucsf.edu> Subject: Re: [chimerax-users] Failing to complete dimer prediction
Hi Tal, Please see these earlier posts about what to try when the calculation is large:
<https://www.rbvi.ucsf.edu/pipermail/chimerax-users/2023-April/005119.html> <https://www.rbvi.ucsf.edu/pipermail/chimerax-users/2023-April/005120.html>
Otherwise, I don't have any suggestions except to say that you can specify a results directory or folder. If any of the intermediate results are written to files (and I do not know if they are), they would appear in this directory.
If you are using a recent version of ChimeraX, the AlphaFold tool will have an Options button, and clicking it will show further settings including "Results directory." For details see the help:
<https://rbvi.ucsf.edu/chimerax/docs/user/tools/alphafold.html> <https://rbvi.ucsf.edu/chimerax/docs/user/tools/alphafold.html#predict>
Best, Elaine ----- Elaine C. Meng, Ph.D. UCSF Chimera(X) team Department of Pharmaceutical Chemistry University of California, San Francisco
On May 1, 2023, at 6:38 AM, Tal Jeffrey via ChimeraX-users <chimerax-users@cgl.ucsf.edu> wrote:
Dear all,
I am trying to use ChimeraX to predict the structure of a dimer as part of my Master’s thesis. I was able to produce a model of the monomeric structure without issue have not been able to do the same for the dimer. I am currently subscribed to Colab Pro so had assumed that the 15.0GB GPU RAM would be sufficient. The structure is 1361 amino acids long but AlphaFold2 is still able to produce the first two models successfully. However, I have not been able to produce any iterations past this stage. Any guidance on how to complete the model would be much appreciated! Alternatively, I would also be glad to extract the output of the first two models (e.g. .pdb files, plDDT plots, PAE plots, MSA heatmaps) but because the prediction does not run to completion I don’t know how to access/extract these.
I have really enjoyed using the program and hope to include the structure of the dimer in my thesis before I submit in three weeks’ time.
Many thanks,
Tal
_______________________________________________ ChimeraX-users mailing list ChimeraX-users@cgl.ucsf.edu Manage subscription: https://www.rbvi.ucsf.edu/mailman/listinfo/chimerax-users
_______________________________________________ ChimeraX-users mailing list ChimeraX-users@cgl.ucsf.edu Manage subscription: https://www.rbvi.ucsf.edu/mailman/listinfo/chimerax-users
Hi Tal, Here are some comments on your questions, although as Elaine says, you should read up on the methods to understand them better. AlphaFold is not like homology modeling. While it can optionally use templates they don't have much impact on a prediction except in cases where AlphaFold has only borderline confidence in its prediction without templates -- this is rare. So most of the AlphaFold web services don't use templates by default since they rarely make a significant difference in the prediction. Sequence coverage plots can include sequence of high and low sequence identity. The main thing AlphaFold needs is enough sequences, usually 100 or more to make a good prediction. pTM and PAE values are entirely different scores and you can read about their definitions align. PAE gives a score for every pair of residues, while pTM (predicted TM) is a single number guessing the agreement between the prediction and unknown true structure. In your attached PAE plots, the mostly blue plot means AlphaFold is confident that all the domains are correctly packed together. In the most red plots I see two different sets of residues that have only red PAE between them, so AlphaFold has no confidence that those two domains are correctly aligned with each other. Tom ChimeraX AlphaFold developer
On May 1, 2023, at 11:13 AM, Tal Jeffrey via ChimeraX-users <chimerax-users@cgl.ucsf.edu> wrote:
Hi Elaine,
Thank you for your response. I kept trying the prediction and fortunately it has worked!
If you don’t mind, may I ask you a few more questions to make sure I am interpreting the confidence metrics correctly? To the best of my knowledge, no experimental data yet exists for the dimer I have modelled. In situations like these, does AlphaFold2 use homology modelling to generate a predicted structure based off of homologous templates? In cases where proteins are predicted without experimental data, are the sequence coverage plots expected to have low sequence identity to query values? How would this change your interpretation of the prediction if the plDDT value was high (~80)? Could you explain how the pTM score relates to the PAE values? Could you explain how you would interpret the PAE plots for predictions of these two different proteins: <image001.png>
<image002.png>
Thank you again and apologies for the bombardment of questions!
Tal
From: Elaine Meng <meng@cgl.ucsf.edu> Date: Monday, 1 May 2023 at 16:22 To: Tal Jeffrey <tal.jeffrey@queens.ox.ac.uk> Cc: chimerax-users@cgl.ucsf.edu <chimerax-users@cgl.ucsf.edu> Subject: Re: [chimerax-users] Failing to complete dimer prediction
Hi Tal, Please see these earlier posts about what to try when the calculation is large:
<https://www.rbvi.ucsf.edu/pipermail/chimerax-users/2023-April/005119.html> <https://www.rbvi.ucsf.edu/pipermail/chimerax-users/2023-April/005120.html>
Otherwise, I don't have any suggestions except to say that you can specify a results directory or folder. If any of the intermediate results are written to files (and I do not know if they are), they would appear in this directory.
If you are using a recent version of ChimeraX, the AlphaFold tool will have an Options button, and clicking it will show further settings including "Results directory." For details see the help:
<https://rbvi.ucsf.edu/chimerax/docs/user/tools/alphafold.html> <https://rbvi.ucsf.edu/chimerax/docs/user/tools/alphafold.html#predict>
Best, Elaine ----- Elaine C. Meng, Ph.D. UCSF Chimera(X) team Department of Pharmaceutical Chemistry University of California, San Francisco
On May 1, 2023, at 6:38 AM, Tal Jeffrey via ChimeraX-users <chimerax-users@cgl.ucsf.edu> wrote:
Dear all,
I am trying to use ChimeraX to predict the structure of a dimer as part of my Master’s thesis. I was able to produce a model of the monomeric structure without issue have not been able to do the same for the dimer. I am currently subscribed to Colab Pro so had assumed that the 15.0GB GPU RAM would be sufficient. The structure is 1361 amino acids long but AlphaFold2 is still able to produce the first two models successfully. However, I have not been able to produce any iterations past this stage. Any guidance on how to complete the model would be much appreciated! Alternatively, I would also be glad to extract the output of the first two models (e.g. .pdb files, plDDT plots, PAE plots, MSA heatmaps) but because the prediction does not run to completion I don’t know how to access/extract these.
I have really enjoyed using the program and hope to include the structure of the dimer in my thesis before I submit in three weeks’ time.
Many thanks,
Tal
_______________________________________________ ChimeraX-users mailing list ChimeraX-users@cgl.ucsf.edu Manage subscription: https://www.rbvi.ucsf.edu/mailman/listinfo/chimerax-users
_______________________________________________ ChimeraX-users mailing list ChimeraX-users@cgl.ucsf.edu Manage subscription: https://www.rbvi.ucsf.edu/mailman/listinfo/chimerax-users
participants (3)
-
Elaine Meng -
Tal Jeffrey -
Tom Goddard