Hi Max,

  ChimeraX AlphaFold prediction uses ColabFold as explained in the ChimeraX documentation

https://www.cgl.ucsf.edu/chimerax/docs/user/tools/alphafold.html#predict

The AlphaFold.ipynb notebook from Google uses a "simplified version of AlphaFold v2.3.2." as described on that notebook web page

https://colab.research.google.com/github/deepmind/alphafold/blob/main/notebooks/AlphaFold.ipynb

and it describes in detail how it was simplified.

  The main difference I think is the deep multiple sequence alignment calculation which ColabFold does on a server in Korea using large TByte sequence databases while Google's AlphaFold notebook instead streams 100 Gbytes of sequence data to the virtual machine for each prediction.  So in general I think the Google notebook will get a poorer sequence alignment.  Also ColabFold runs much faster often 5 times faster.

Tom




On Apr 18, 2024, at 12:37 PM, Coyle, Maxwell C via ChimeraX-users <chimerax-users@cgl.ucsf.edu> wrote:

Hello!

I’ve been predicting structures recently using the AlphaFold tool through ChimeraX (which directs to the alphafold21_predict_colab.ipynb notebook), but I’ve noticed I get often quite different results when I run the same prediction through the AlphaFold.ipynb notebook directly in Colab. I was wondering if anybody could point me to the major differences between these two implementations? Are they searching different libraries to generate a multiple sequence alignment? Are they using different numbers of iterations or energy minimization strategies?

Thanks for any guidance,
Max

Max Coyle, PhD
Postdoctoral Fellow, Bellono Lab
Dept of Molecular and Cellular Biology
Harvard University
mcoyle@fas.harvard.edu




_______________________________________________
ChimeraX-users mailing list -- chimerax-users@cgl.ucsf.edu
To unsubscribe send an email to chimerax-users-leave@cgl.ucsf.edu
Archives: https://mail.cgl.ucsf.edu/mailman/archives/list/chimerax-users@cgl.ucsf.edu/