Modeling updates

[schallerdavid]

Description

This PR adds important updates, new functionalities and tests to the KinoML modeling pipeline.

Todos

Notable points that this PR has either accomplished or will accomplish.

• [x] remove temporary fixes for loop modeling
• [x] switch to new RCSB API
• [x] implement macrocycle omega
• [x] switch from klifs_utils to opencadd
• [x] support uniprot_id as kinase_identifier
• [x] improve superposing by allowing to provide a list of residues to align on
• [x] mutation to standard amino acids in OEKLIFSKinaseHybridDockingFeaturizer
• [x] pocket similarity instead of identity for template selection in OEKLIFSKinaseHybridDockingFeaturizer
• [x] remove expression tags from pdb structures
• [x] custom conformation to OEKLIFSKinaseHybridDockingFeaturizer
• [x] implement structure manipulation pipeline: deletion -> insertion -> mutation
• [x] custom sequence to OEKLIFSKinaseHybridDockingFeaturizer
• [x] implement OEKLIFSKinaseApoFeaturizer

Status

• [ ] Ready to go

[review-notebook-app[bot]]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[schallerdavid]

Hi @t-kimber and @jaimergp ,

I merged master into this branch. This branch adds several improvements to the structural modeling pipeline and additionally adds tests. Pytest only reports a single failure, but this is intended since it is simply not implemented yet. My complex featurizers seem to work. Is there anything else I can test or you want to test before I merge this into master?

Cheers, David

[jaimergp]

"Pipeline" tests are done in https://github.com/openkinome/experiments-binding-affinity. Talia should be able to tell you how to add your featurizers to some examples there and run it locally!

[codecov-commenter]

Codecov Report

Merging #34 (2889cca) into master (4b24e11) will increase coverage by 11.73%. The diff coverage is 46.63%.

[schallerdavid]

Except for test_access_by_index_roundtrip all tests are passing. But this is not implemented yet. The featurization tests in experiments-binding-affinity are passing but not the model training. However, this was already the case with the current master (see #13). Merging now.