Alphafold prediction: invalid character in the sequence

[Tom Goddard]

ChimeraX Alphafold prediction stopped working I believe on March 14, 2024 as described in these 3 bug reports #14771, #14775, #14776.

This was caused by a change in the Google Colab web interface which made sending the sequence from ChimeraX to Google Colab fail silently. Then it tried to run the prediction on the default sequence text which was “Paste a sequence …” containing spaces that led the invalid character error.

Sending the sequence was done in a very hacky way because there are no Colab APIs to do this. So ChimeraX just set the sequence using Javascript to fill the sequence entry field. Colab changed that entry field from an html panel-input element so the entry field is no longer found.

[Tom Goddard]

Predictions first check for an alphafold status page on the rbvi web server to see if it is down. I added that page so users are told that prediction is broken and it may be a few days or more before fixed.

[Tom Goddard]

Ideally I’d like to fix this in the Colab python notebook so users do not need to get a new ChimeraX. There is a tricky way to do this. I can make the notebook create an html entry field using a panel-input element in the cell output so that ChimeraX will send the sequence to that entry. And I can put a change callback on that entry which calls back into Python to set the Python sequence variable.

There are some tricks to make this work. Some notebook python code must run to register the change callback and ChimeraX does not run the notebook until after setting the sequence, so currently no python is run before the sequence is set. But there is an option to run the first cell of a notebook automatically when it is loaded. But ChimeraX runs the first cell to do the prediction. I can put in code in the first cell that registers the change callback on the first execution and on the second runs the prediction.

Not sure this will work, but I tested the pieces in a notebook without alphafold or chimerax that worked. Difficult to try on my iphone from China.

This did not work. It appears that ChimeraX gets the page loaded callback before the first cell runs so it tries to send the sequences before the panel-input is created. Also the cell is only run once, possibly because ChimeraX tries to press the run button while it is already running.

[Tom Goddard]

Another approach is to just fix the sequence sending in the daily build. It appears that the panel-input entry field was changed to an md-outlined-text-field.

[Eric Pettersen]

If the status-checking mechanism were sophisticated enough to discriminate based on the ChimeraX version, once you fix this you could present a message to older versions to upgrade to the current version, and have the current version just work with no message.

[Tom Goddard]

If I fix Alphafold prediction with a new ChimeraX I will have the new version check a new status file and update the status file that all older ChimeraX versions check to say a new ChimeraX is needed to run Alphafold.

[Eric Pettersen]

Sounds like a simple effective approach.

[Tom Goddard]

Fixed in daily build. Made ChimeraX set md-outlined-text-field entry. This is a pretty bad solution since it will break again when Colab changes its web site. Also bad because Alphafold prediction will remain broken in all older ChimeraX. But best I could do working from China on iphone. Will revisit next month.

[Tom Goddard]

Should make the Javascript that sets the sequences and that runs the notebook ChimeraX settings that can be updated from a json file on the rbvi web server if those javascript calls fail. This will allow fixing alphafold prediction when the Colab web page changes. The ChimeraX Alphafold database fetch code already does similar runtime updates.

[Tom Goddard]

Updated the alphafold prediction error message shown by older ChimeraX to say prediction is fixed in the daily build.