fix: check PSSMs

[DaniBodor]

Summary:

• Change the assumed order in which the pssm file is read to alphabetical by 3-letter-code (was 1-letter-code, but this did not match pssm files used).
  • Perhaps would be even nicer to read the order from the top of the pssm file instead of assuming a fixed order in the future.
• Include sanity check on matching amino acid order of pssm and pdb file
  • Test for both correct aa order (test_incorrect_pssm_order) and missing entries (test_incomplete_pssm) in pssm in test_query.py
• Test that not having a pssm file throws an error if conservation module is selected
  • Improve the error message when no pssm file is found
• Change default feature modules in query.process to only components and contact
  • Option to use 'all' as input, which will use all existing feature modules
• Improvements to test_querycollection.py
  • Create function to check for expected presence/absence of features
  • Make separate tests for all features and default feature modules
  • Other minor improvements
• Improve test speed, e.g. by reducing the number of queries and epochs in some of the tests (as this PR was on hold for a while, that may have been implemented by some other PR in the meantime).
• Fixed pytorch 2.0.0 in the workflow, because no wheel is available for 2.0.1. This may cause problems in the future as well if pytorch and the wheels are not updated synchronously.

Note: It would be nice if there were some form of documentation on what the pssm files should look like and/or where and how to build it correctly, as it is very easy get it wrong and very difficult to diagnose or even detect if it is. This is not implemented in this PR.

[github-actions]

This PR is stale because it has been open for 14 days with no activity.

[gcroci2]

Nice checks :-)

I'd avoid adding exclude_feature_module (the least parameters the better), and I'd implement the solution we were discussing, which is defaulting the modules to the essentials and making feature_modules required. I don't think opening a new PR for this is needed, right?

[gcroci2]

Great! I've also just noticed that in one of my branches. Also I had a question: why is the line 66 needed? (# Install optional pyg dependencies) Shouldn't conda install pyg -c pyg take care of the pyg dependencies?

[DaniBodor]

Don't know why, but it doesn't. I guess because they are optional?

[gcroci2]

With conda, from their website it should not be necessary. Only with pip. So if it doesn't work using only conda install pyg -c pyg their docs are wrong

[gcroci2]

I added comments about minor edits in the process method. I think that explicitly defaulting the modules (not using None) and allowing for 'all' but not None is clear from the user perspective.

[DaniBodor]

I added comments about minor edits in the process method. I think that explicitly defaulting the modules (not using None) and allowing for 'all' but not None is clear from the user perspective.

Agreed!