v3.2.0

New:

• Added explain module. This module takes any CAPICE model, using v3.0.0 or greater, and exports the feature importance of said model with importances: gain, total_gain, weight, cover and total_cover, ordered by gain.
• Added -t/--threads flag to train to specify how many computing cores can be used during training.

Changes:

• Changed behaviour of VEP processor consequence.py. Added 3 missing possible consequences that could be within the output of VEP but were not processed yet. The processor will now also throw a warning if it encounters a consequence it does not support.
• Code refactoring to main_train.py
• Changed feedback to user for exports. Now if a user supplies an output argument that is a full path including filename, it will not print where the file is placed. In any other case, such as no supplied output, only a path as output or only a filename as output, CAPICE will now print where the file is placed. This change is implemented for predict, train and explain.
• Increased strictness of the output argument. CAPICE will no longer add a .gz extension to the output, it is required to be supplied by the user.
• Fixed data types errors that would only be raised on Windows machines.
• Fixed the VEP command.

v3.1.2

Changes:

• Bumped Numpy to version 1.22.0 due to security vulnerability.

v3.1.1

New:

• Added standalone version validator that checks both Semantic and PEP440 versions for both CAPICE and the model. Also checks if the versions are compatible.

Changes:

• Fixed a bug where CAPICE would error on file extensions when output argument -o . was supplied.
• Fixed code smell within args_handler_predict.py.
• Fixed a bug that didn't check for required input feature feature_type for predictions.
• Fixed a bug where the PoC test resources model was considered out-of-date.

v3.1.0

Changes:

• Improved version checking for CAPICE models. CAPICE models now will work if a patch or minor version mismatch is present, but will error when a major or release candidate mismatch is present.
• #95 README updates
• #96 Move __version__ to __init__ instead of __version__.py

v3.0.0

New features

• New Command Line Interface with improved separation between the predict and train module.
• Customized XGBoost model that includes impute values and CAPICE versioning.
• Added support for edge cases (breakends, symbolic alleles and alternative contigs).
• Revamped input that no longer depends on CADD 1.4 but instead depends on VEP.
• Added a manual annotator module that processes the VEP features. Each module is separate and are loaded in dynamically, no codebase changes have to be made in order to start supporting new features.
• Added bash script to convert a VEP output VCF to a CAPICE ready input TSV.
• Added helper script to balance out an input dataset on allele frequency, consequence and the amount of benign vs pathogenic samples.
• Added Travis CI for automated testing and deployment.
• Created access point capice.
• Added support for Python 3.10.
• Added input requirement "SpliceAI".

Changed

• Separated CAPICE framework from CAPICE model. New models can now be released independently.
• Variants are no longer removed if they contain no feature at all. Instead, CAPICE will now throw an error when a gap is present within the CHROM, POS, REF or ALT column.
• Reduced overall code complexity.
• PEP8 formatting guidelines with maximum line length 100 characters.
• Updated testing.
• Changed logging. It is no longer possible to let CAPICE create a logfile. Instead, CAPICE logs DEBUG and INFO to system STDout and WARNING, ERROR and CRITICAL to STDerr.
• Changed logging output format to YYYY-MM-DD <loglevel>: <msg>
• Changed code structure of CAPICE to follow the MOLGENIS Python template.

Removed

• Removed deprecated files in CAPICE_example.
• Removed ability to create train-test / validation datasets directly through CAPICE. Use the supplied helper script in order to perform these steps.
• Removed support for Python <=3.7.

Bug fixes

• Fixed a bug when -v / --verbose flag was called, CAPICE would not output DEBUG level logmessages.
• Fixed a bug within training that would throw a ValueError when used with XGBoost 1.4.2

v3.0.0-rc4: Merge pull request #88 from molgenis/rc4_spliceai

Release Candidate 4: SpliceAI

v3.0.0-rc3: Merge pull request #68 from molgenis/python3.10

feat: Python3.10 support

v3.0.0-rc2: Merge pull request #85 from molgenis/fix/github-releases-assets

Fix Travis GitHub releases deploy: do not upload assets

v3.0.0-rc1

Changed

• Fixed a bug where the set entry point for CAPICE resulted in
  "ModuleNotFoundError: no module named capice"
• Changed code structure to follow MOLGENIS Python template
• Bumped minimally supported python version to 3.8

v3.0.0-beta2

New:

• Added VEP variant Consequence processor to better differentiate the consequence of a variant, potentially increasing model performance for in particularly pathogenic variants.

Changed:

• Changed the version within setup.py to import the version from the project instead of duplicate define.