[chimerax-users] Error reported by AlphaFold

[Tom Goddard]

Hi Juntao,

  The error is not related to whether your sequences form a complex.  It is simply a bug or limitation of AlphaFold that is causing the failure.  There are many bugs in AlphaFold that can cause this.  To figure out which one you are hitting you should look in the log file that the error message lists.  If ChimeraX downloaded the results (I think it will even if all 5 predictions give errors) then they will be on your computer in

	~/Downloads/ChimeraX/AlphaFold/prediction_1

and the errors for each prediction will be in files model_1_error, model_2_error...

  If you send that error log file I can probably tell you what the cause is.  Sometimes it is CUDA is out of memory (GPU does not have enough memory) -- common if your total sequence length is > 1000 amino acids.  But other failures happen, e.g. in energy minimization.  If the problem is the sequence length is too long, then you should try to run AlphaFold on your own computers with a modern GPU with lots of memory instead of Google Colab which uses 5 year old GPUs with only 16 Gbytes of memory.

  Here are some test runs of AlphaFold giving you some examples of the kinds of failures that occur.

	https://www.rbvi.ucsf.edu/chimerax/data/alphafold-jan2022/afspeed.html

  Tom



> On Feb 14, 2022, at 10:09 AM, Juntao Yu via ChimeraX-users <chimerax-users at cgl.ucsf.edu> wrote:
> 
> Dear ChimeraX community,
> 
> I am a graduate student from Harvard Medical School who recently became interested in using AlphaFold-Multimer to predict protein complex structures of my interest. I have been tried a few times of predicting one protein complex and I was able to finish all MSA. However, when AlphaFold starts predicting the structure, all five models were reported as “error”. It is surprising to me because it is known that these two proteins should form a complex. I noticed in the script on the line is: “print ('\nAlphaFold generated an error computing %s, error logged to %s\n'% (model_name, error_log_path))”. The results printed exactly what this line says. 
> 
> I am wondering whether it means the Multimer is not able to predict this complex structure? Is there any tutorial that I can find to change parameters in order to try to improve and get a structure?
> 
> Thanks,
> Juntao
> 
> ------------------------------------------
> Juntao Yu | Ph.D. candidate
> Moazed Lab
> Department of Cell Biology
> Harvard Medical School
> 240 Longwood Avenue
> Boston, MA 02115
> 
> _______________________________________________
> ChimeraX-users mailing list
> ChimeraX-users at cgl.ucsf.edu
> Manage subscription:
> https://www.rbvi.ucsf.edu/mailman/listinfo/chimerax-users

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.rbvi.ucsf.edu/pipermail/chimerax-users/attachments/20220214/fd4ce63a/attachment.html>