ChimeraX ColabFold bug

[jjbeck@…]

Hello,

I have been using ChimeraX to operate ColabFold for a few years now, but suddenly I am getting the same runtime error no matter what sequence I attempt to upload. I have attached it to this email. Please let me know if you have encountered this error yourselves and if so what I can do to remedy. This has happened to me with verions 1.4, 1.5, and 1.6rc. Any help is hugely appreciated.
[cid:af34d662-ff8b-4643-9f5b-32b9553f7635]
Thank you,
Jordan Beck

[jjbeck@…]

Added by email2trac

[Eric Pettersen]

Reported by Jordan Jeffrey Beck

[goddard@…]

Thanks for reporting the problem. The error message says that installing ColabFold on the Google Colab machine is failing. 

Eric or Jordan could you run an Alphafold prediction and when it gives this error click on the folder icon at the left edge of the Alphafold Run window and download the install_log.txt file so we can see what the install error is. Also get the install_colabfold.sh file so we can see the actual install command. Attach those 2 files to this ticket. 

I am in China and have only a phone so I do not have a way to test or fix but possibly the error message will suggest a simple fix. 

ChimeraX is trying to install a specific older version of Colabfold, so I am surprised it suddenly started failing since no new Colabfold changes should effect it. 

[Tom Goddard]

ChimeraX Alphafold prediction is no longer working because Google Colab updated today from Python 3.9 to 3.10. The version of ColabFold ChimeraX is using requires tensorflow-cpu version < 2.8 and that is not available for Python 3.10. The fix is to update ChimeraX to use a newer version of Colabfold. Still considering how to do that since I am away from the office with no computer for a few weeks.

A work-around is to tell Google Colab to use Python 3.9 using the “command pallete” icon in the AlphaFold Run panel (lower left corner rectangle icon with 2 horizontal lines) and choose Use Fallback Runtime Version as described here

​https://medium.com/google-colab/colab-updated-to-python-3-10-27eb02daa162

Then press the red triangle icon near the top left of the AlphaFold Run panel to restart the run with Python 3.9.

[Tom Goddard]

Same ChimeraX Alphafold error also reported as github alphafold issue

​https://github.com/deepmind/alphafold/issues/755

[Tom Goddard]

Partly fixed.

I updated ColabFold to v1.5.2 which is the most recent version from March 2023 but that failed to install in Colab Python 3.10.11 because the required Python version was incorrectly specified in the ColabFold pyproject.toml file. But another commit 1 week later fixed that. So I used that unversioned Colabfold. Monomer and dimer predictions of 1aom worked (although the structures were bad, and I think they were good with previous alphafold versions).

This newer Colabfold is also using Alphafold 2.3.1 which is supposed to handle larger structures and has new multimer networks.

Unfortunately OpenMM 7.5.1 used for energy minimization fails to install in Python 3.10 with dependency inconsistencies. Energy minimization is off by default in ChimeraX alphafold predictions, but there is an option to enable it. The fix is probably to update the colab script to instead install openmm 8.0.0. Not sure if that will work without further changes. I think openmm was moved out of the simtk directory. Will have to test.

Did the colabfold update from China over vpn and vnc from an ipad — excruciatingly slow. VPN has become unusably slow (1 screen update per minute). Could be the great firewall is onto the rbvi vpn. So Openmm fix may need to wait a few weeks until I am back.