Multiprocessing for structural modeling

[schallerdavid]

Description

Add support for multiprocessing for KLIFS complex featurizers. This PR will also alter abstract featurizer classes in kinoml/features/core and subclasses thereof.

Todos

• [x] add _pre_featurize method to BaseFeaturizer
• [x] move KLIFS queries to _pre_featurize method of KLIFS complex featurizers
• [x] KLIFS complex featurizers return empty MDAnalysis universes in case no valid kinase structure is available to model the given system
• [x] move multiprocessing arguments to __init__ of BaseFeaturizer
  • [x] default use_multiprocessing=False
  • [x] if use_multiprocessing=True, use all available threads
  • [x] n_processes argument to select number of threads
• [x] adapt all classes in kinoml/features/core to new multiprocessing handling
• [x] add arguments for custom cache and output directory
• [x] allow keeping the protonation state of the molecule to dock
• [x] implement multiprocessing handling via dask

Other improvements

• [x] add support to select specific PDB structures for modeling in OEKLIFSKinaseHybridDockingFeaturizer

Status

• [ ] Ready to go

[codecov-commenter]

Codecov Report

Merging #52 (f75fba4) into master (6500257) will increase coverage by 0.17%. The diff coverage is 31.69%.

[schallerdavid]

Hi @t-kimber , everything seems to work out with the different parallelization schemes. I also introduced a few changes to ligand featurizers because of the changes I made to the basefeaturizers. Do you wanna give it a try or have any suggestions before I merge?