Skip to content

v4.0.0

Choose a tag to compare

View all tags
@ArthurDeclercq ArthurDeclercq released this 17 Jul 20:49
· 22 commits to main since this release
v4.0.0
93e6125

Added

  • New rescoring engine: ristretto, a lean, dependency-light (numpy/scikit-learn/pandas only) reimplementation of the Percolator/Käll semi-supervised algorithm, purpose-built for MS²Rescore.
  • New rescoring configuration option: train_fdr and model ("svm", default, or "lda", faster but less powerful).
  • New top-level report_fdr configuration option: FDR threshold used for console-logged identification counts, the HTML report's stats/charts, and FlashLFQ output filtering. Previously hardcoded at 1% throughout.
  • Automatic inference of search-engine score direction (higher-is-better vs. lower-is-better) via spectrum-competed target-decoy evaluation, replacing the user-set lower_score_is_better option. Grouped by run, so multi-file input sharing native spectrum/scan IDs across runs doesn't corrupt the inferred direction.
  • Rescoring result tables (<prefix>.psms.tsv, .peptidoforms.tsv, .peptides.tsv, .proteins.tsv, .weights.tsv) are now always written as plain TSV.
  • ms2rescore-report CLI: new --fdr option to regenerate a report at a different FDR threshold without rerunning rescoring.
  • New MS2 feature generator using Rust-based ms2rescore_rs for direct spectrum feature extraction (intensity ratios, matched ion counts/percentages, hyperscore).
  • New basic features: theoretical_mass, experimental_mass, mass_error, pep_len.
  • annotate_spectra(): annotates all PSM spectra once before feature generators run, eliminating redundant per-generator spectrum parsing.
  • Top-level configuration options fragmentation_model, tolerance_value, and tolerance_mode to control centralized fragment ion annotation (defaults: cidhcd, 0.02 Da).
  • Mumble integration (optional, beta): a new PSM generator for exploring alternative peptide identifications via candidate mass-shift modifications (pip install ms2rescore[mumble]).
  • Intermediate file output on feature-generation or rescoring errors, enabling recovery by rerunning with a modified configuration instead of restarting from scratch.
  • Feature generators are intelligently skipped when all their features are already present in the input PSM file (e.g., on a recovery run).
  • Standalone HTML report regeneration from a PSM TSV file alone -- no config or log file required; before/after comparisons are reconstructed from the PSM list's provenance data.
  • ParseSpectrumError exception for spectrum-parsing failures.

Changed

  • MS2 and MS2PIP feature calculation migrated to Rust via ms2rescore_rs (~5x speed-up).
  • Spectrum files are parsed and annotated once, up front, and shared across all feature generators.
  • DeepLC upgraded to its v4 API: dataset-wide processing with per-run calibration or finetuning. New multitask model leads to much improved performance, even without finetuning.
  • IM2Deep upgraded to its v2 API (im2deep>=2.0.1): dataset-wide processing with per-run CCS calibration using reference peptides.
  • Basic feature generator uses fixed charge encoding (charges 1-6) instead of a dynamic per-dataset range.
  • HTML report generation (in-run and standalone) reconstructs before/after rescoring comparisons from the main PSM list's provenance data, rather than relying on separately persisted result tables.
  • Report/identification-overlap comparisons key on (run, spectrum_id) instead of bare spectrum_id, so multiple input files reusing the same native spectrum IDs no longer collide.
  • Multi-run PSM lists are disambiguated during rescoring/competition via a run identifier, instead of relying on spectrum ID alone.
  • max_psm_rank_output > 1 now applies consistently across the main output, rescoring tables, and report: multiple ranked PSMs per spectrum, with q-values/PEPs computed per-row rather than through full spectrum competition. Intended for surfacing ambiguous candidates (e.g. from Mumble), not a statistically rigorous FDR-controlled count.
  • Main PSM list output renamed <prefix>.psms.tsv → <prefix>.tsv; the crash-recovery intermediate file renamed the same way (<prefix>.intermediate.tsv).
  • Dependencies upgraded: deeplc>=4.0.0, im2deep>=2.0.1, ms2pip>=4.2.0, ms2rescore_rs>=0.5.0. Added pyarrow.
  • numpy 2.0 compatibility.
  • Python 3.11 or newer is now required.

Removed

  • [BREAKING] Mokapot rescoring engine and dependency removed.
  • [BREAKING] Percolator CLI integration removed (the separate engine that shelled out to a locally-installed percolator binary). ristretto is now the only rescoring engine.
  • [BREAKING] MaxQuant feature generator removed; functionality consolidated into the MS2 feature generator.
  • [BREAKING] ionmob feature generator removed; replaced by IM2Deep v2.
  • [BREAKING] rescoring_engine configuration option removed (mokapot/Percolator-specific: fasta_file, write_weights, write_txt, protein_kwargs), replaced by rescoring.
  • [BREAKING] Top-level fasta_file configuration option and FASTA-based protein inference removed (mokapot-specific). Does not affect Mumble's separate psm_generator.mumble.fasta_file option, which is unrelated and unchanged.
  • [BREAKING] lower_score_is_better configuration option removed; score direction is always auto-inferred now, with no configuration-level override.
  • [BREAKING] write_rescoring_tables configuration option removed -- rescoring tables are unconditionally written.
  • [BREAKING] PIN (Percolator) file output removed; the main PSM list TSV already carries all rescoring features.
  • [BREAKING] Ability to skip rescoring via configuration removed; rescoring always runs.
  • [BREAKING] ms2_tolerance, spectrum_path, and spectrum_id_pattern parameters removed from MS2PIPFeatureGenerator. Fragment mass tolerance is set globally via tolerance_value / tolerance_mode.
  • [BREAKING] spectrum_path, spectrum_id_pattern, mass_mode, and processes parameters removed from MS2FeatureGenerator. Spectra are provided via centralized annotate_spectra().
  • [BREAKING] ms2rescore.utils (public Python API) renamed and split into two internal modules, ms2rescore._utils and ms2rescore._ristretto_utils -- neither is part of the public API.
  • deeplcretrainer dependency removed (functionality merged into DeepLC v4).
  • tomli dependency removed (only required for Python <3.11).

Fixed

  • processes=-1 (ms2rescore default) passed to DeepLC num_threads, which requires a positive integer or None.
  • Q-value NaN check in parse_psms.py failed when qvalue array contained None values.
  • BrokenExecutor not caught during mokapot rescoring, producing unclear crashes on worker failure.
  • GUI runs never wrote an HTML log file, unlike CLI runs.
  • Out-of-memory errors from multiprocessing during spectrum parsing.

Full Changelog: v3.2.1...v4.0.0

Assets 3
Loading