v1.0.1

Hotfixes for missing packages on installation.
Improves reproducibility using constraints.txt and requirements.txt. Installation now runs via pyproject.toml (instead of setup.py).

v1.0.0

Full open source v1.0.0 release.
This includes a new MS/MS prediction model with a focus on long-range interactions. It includes the following features:

• Trained on large MSnLib v7 library (24741 trainings compounds after curation; 3092 for validation; 3092 for testing)
• Residual connections (ResNet) - optional
• GNN Layer stacking (concatenation of intermediate GNN embeddings)
• Subgraph (fragment) features
  • element composition
  • subgraph embedding (resepctive node pooling)
• KL Divergence Loss function
• Layer normalization
• adjustable pooling functions (max and avg)
• Higher regularisation (e.g. Weight decay, dropout etc.)
• Improved feature and target specification
• Modularization and refactoring of the FioraModel code base
• Expanded gridsearch options
• Metabolite indexing for faster training and testing
• Version controlling (with backward compatibility)
• Sorted output peaks
• Adjusted help and info messages
• Various small bugfixes and adjustments

v0.1.2

Quality of Life Improvements

• Fix issues with MSP output format
• Fix small deviation in hydrogen weight calculation
• Add more information to MGF and MSP output formats
• Add a debugging command line option (for detecting invalid smiles)
• Add an option to set a minimum peak probability threshold
• Update the README.md with images and a reference to the publication

v0.1.1

Small updates and stability patches:

• Extended unit tests for usability after installation
• Improved training procedure
• Automatic tracking of additional statistics and scores during spectral prediction
• Improved visualization and plots
• Several hotfixes

v0.1.0

Initial Release

Major features:

• Release of (first) version-controlled open-source model weights (Fiora OS v0.1.0)
• Installation guide and unit tests (thoroughly tested)
• Jupyter notebooks with training, testing, and live prediction capabilities
• Various Python modules for metabolomics, machine learning, tandem mass spectrometry and visualizations