v4.2.0-beta.1

MS²PIP 4.2.0 replaces the C/Cython backend with a Rust engine, cutting prediction time by ~30% (119s → 84s on a real dataset) and liminating a scaling bottleneck that made more CPU cores slower. Predictions and spectrum annotation are now handled in a singleparallelized Rust call. The package is now a pure-Python wheel — no compilation on install — and supports Python 3.11+.

---

Added

• All prediction, m/z calculation, and spectrum annotation now use ms2rescore-rs (Rust); replaces C/Cython backend
• ms2_tolerance_mode parameter ("Da" or "ppm") on all correlation/annotation functions and CLI
• correlate now accepts preloaded spectra on PSMs (MS2Spectrum / AnnotatedMS2Spectrum)
• read_psms accepts list[PSM] in addition to PSMList, str, and Path
• Replaced remaining C-based model files with native XGBoost models (HCD2019, TMT, and HCDch2)
• XGBoost models are now pre-loaded across predict_library batches for improved efficiency
• rt and im optional dependency groups for DeepLC and IM2Deep
• Min/max length validators on ProteomeSearchSpace
• CUDA_VISIBLE_DEVICES workaround on XGBoost model load (dmlc/xgboost#11283)
• API docs for ms2pip.result and ms2pip.spectrum

Changed

• Minimum Python version raised to 3.11
• Package is now a pure-Python wheel (no compiled extensions)
• Batch processing uses Rayon-parallelized Rust calls instead of Python multiprocessing
• DeepLC integration uses deeplc.predict() functional API (replaces class-based DeepLC wrapper)
• IM2Deep integration uses im2deep.predict() functional API (replaces legacy im2deep.im2deep.predict_ccs)
• predict_library now runs RT/IM predictions once on the full filtered PSMList before batching (closes #243)
• Invalid peptidoforms (unsupported amino acids, length outside 4–100, missing charge) are skipped with a summarized warning instead of raising
• Bumped ms2rescore-rs requirement to >=0.5.0a3,<2
• Bumped psm_utils requirement to >=1.5

Removed

• C/Cython components: _cython_modules/, _models_c/, setup.py, MANIFEST.in
• All Python-based multiprocessing (_Parallelized class)
• _utils/retention_time.py and _utils/ion_mobility.py (inlined into core.py)
• iRT calibration peptides (DeepLC v4 handles calibration internally)
• Minor breaking change: ms2pip.constants.MODELS entries no longer contain id, peaks_version, or features_version keys (C/Cython routing metadata); replaced by fragmentation. Code accessing these keys directly will break.

Fixed

• annotate-spectra CLI crash (.with_suffix() called with no argument)
• write_correlations now accepts Path objects
• _peptidoform_spaces empty check uses is None instead of truthiness

---

All changes: v4.1.2...v4.2.0-beta.1

4.2.0-alpha.3

Full Changelog: v4.2.0-alpha.2...v4.2.0-alpha.3

v4.2.0-alpha.2

Full Changelog: v4.2.0-alpha.1...v4.2.0-alpha.2

v4.2.0-alpha.1

Full Changelog: v4.2.0-alpha.0...v4.2.0-alpha.1

v4.2.0-alpha.0

Full Changelog: v4.1.2...v4.2.0-alpha.0

v4.1.2

Fixed

• Prefer (faster) Genesis for model downloads, with a fall back to Zenodo
• CI: Update build runners for macOS (see actions/runner-images#13046)
• Fix logging for spectral library prediction (by @paretje in #257)

---

Full Changelog: v4.1.1...v4.1.2

v4.1.1

Fixed

• ⬆️ Fix support for sqlalchemy v2, keeping backwards compatibility with v1.4 (#249, fixes #250)
• 📝 Fix typo of max_length in search_space documentation (#245 by @paretje)
• 👷 In CI workflows, don't build dependencies like pyarrow from source. This can result in failed build workflows.

---

Full Changelog: v4.1.0...v4.1.1

v4.1.0

Added

• ✨ Support for Thermo raw (requiring dotnet runtime) and gzipped spectrum files, through mobiusklein/mzdata (#226, by @paretje)
• ✨ New Python API usage mode correlate-single to correlate to a single ObservedSpectrum object with predictions (#232)
• ✨ Added support for Python 3.12 and 3.13, Numpy v2, and newer Pandas and XGBoost versions (#228)
• ✅ CI: Add integration test for predict-single (#228)

Changed

• ⚠️ Ion mobility is now returned instead of collisional cross section when IM2Deep is used through the add_ion_mobility option (#236, by rodvrees)
• 🏗️ Model files are now downloaded from Zenodo instead of Genesis (#225, by paretje, fixes #229)
• 📝 Docs: Updated README to include correlate-single; moved webserver API docs from old Wiki (#239)

Removed

• 🐍 Removed support for Python 3.8 (EOL) and for musllinux distributions (#228, #237)
• 🗑️ Removed old conversion scripts, mostly implemented in psm_utils (#239)
• 🗑️ Removed old train script, to be replaced with page in docs (#239)

v4.0.0

Fully refactored and drastically more user-friendly version of MS²PIP:

• More modular Python API
• One consolidated command-line interface with subcommands
• Support for all file formats readable by psm_utils
• Support for MGF, mzML, and Bruker raw spectrum files
• Support for ProForma 2.0 peptide notation (no modification configuration required anymore)
• Support for multiple peptides/PSMs per spectrum
• Detailed documentation on ms2pip.readthedocs.io
• ...

v3.13.0

Added

• New timsTOF 2024 model added (now alias for timsTOF) by @ArthurDeclercq in #211

Full Changelog: v3.12.0...v3.13.0