Pyopenms

PyOpenMS

Overview

PyOpenMS provides Python bindings to the OpenMS library for computational mass spectrometry, enabling analysis of proteomics and metabolomics data. Use it to read/write MS file formats, process raw spectra, detect and quantify features, identify peptides and proteins, and run end-to-end LC-MS/MS pipelines.

This skill ships ready-to-run scripts in scripts/ covering the most common high-level workflows. Prefer running a script over writing new code—each is a parameterized CLI tool that handles loading, processing, and export. Drop into the Python API (and the references/) only when no script fits.

Installation

uv venv --python 3.13
uv pip install "pyopenms==3.6.0" pandas numpy matplotlib

Verify (note: __version__ works, but the bundled binary prints a one-line memory-status notice on import that is harmless):

import pyopenms as ms
print(ms.__version__)  # 3.6.0

Scripts (start here)

Run with python scripts/<name>.py --help for full options. Input formats differ by script; inspect its help. Commands below run from the skill directory with the environment activated. Native regression tests use tiny synthetic files; instrument-specific detection/search performance is not validated.

Inspect & convert

Feature detection & quantification

Annotation

Identification

Chemistry

Targeted & visualization

Common script recipes

# Inspect a file
python scripts/inspect_ms_data.py sample.mzML --spectra-csv spectra.csv

# Untargeted metabolomics: features for one sample
python scripts/detect_features_metabo.py sample.mzML --out-csv features.csv

# Full multi-sample quantification study
python scripts/align_link_quantify.py s1.mzML s2.mzML s3.mzML --out-prefix study
python scripts/consensus_to_matrix.py study.consensusXML --out quant.csv --normalize median

# Peptide chemistry
python scripts/mass_calculator.py --peptide "PEPTIDEM(Oxidation)K" --charges 1 2 3 --isotopes 5
python scripts/digest_protein.py proteins.fasta --enzyme Trypsin --missed 2 --out peptides.csv

# Identification post-processing
python scripts/process_identifications.py search.idXML --fasta db.fasta --fdr 0.01 --out filtered.idXML --csv hits.csv

Identification confidence

--fdr estimates top-hit PSM q-values from one comparable search run and rejects missing labels, non-finite scores, mixed score types/directions, and absent target or decoy top hits. It does not recalibrate existing q-values. Before using it, verify target/decoy annotations, score direction, and the search database used to generate the hits. The script applies FalseDiscoveryRate to peptide identifications; its threshold does not establish protein-level FDR. Report the tested unit (PSM, unique peptide, or protein), pooling/search settings, decoy strategy, and threshold explicitly. Protein inference and protein-level error control need their own validated workflow; do not label all inferred proteins “1% FDR” from the peptide-hit filter alone. See the OpenMS FDR API.

Version 3.6.0 API and scientific checks

OpenMS 3.6.0 moved pyOpenMS to nanobind. The old documentation site's latest page still identifies itself as 3.5.0dev; use installed help() and release source when a signature disagrees. This skill's version 3.0 updates the binding calls and changes the peptide detector option from --mz-tol-ppm to --mz-tol-da: its native algorithm uses an absolute m/z tolerance, so conversion at an arbitrary m/z 400 was incorrect for the rest of the mass range.

• MassTraceDetection.run(exp, 0) returns traces; ElutionPeakDetection.detectPeaks(traces) returns split traces; FeatureFindingMetabo.run(traces, features) fills the map and returns a tuple. It no longer accepts a third chromatogram output list.
• Param.keys() returns strings. Use PeptideIdentificationList for mutable IDs; IdXMLFile.load(path) also supports returning (proteins, peptides) in 3.6.
• Feature tables use rt/mz; consensus tables use get_intensity_df() and get_metadata_df(). RT and chromatogram time are seconds; m/z is Th; neutral mass is Da. An OpenMS option named Da on an m/z window is absolute m/z tolerance. Record whether a ppm tolerance is a half-window (the XIC script uses abs(observed-target) <= target*ppm/1e6).
• Check spec.getType() against SpectrumSettings.SpectrumType; sorted m/z says nothing about centroid/profile status. Detectors require centroided MS1 and exclude MS2. Unknown type needs a justified --assume-centroided; smoothing or picking unknown type needs --assume-profile. Do not peak-pick centroid data.
• process_spectra.py --ms-level scopes every operation to that level and keeps chromatograms unchanged. Within-spectrum normalization changes quantitative signal and is usually inappropriate before label-free intensity comparison.
• Charge zero means unknown. The mass calculator's positive charge magnitudes assume protonation/deprotonation only; sodium, ammonium, multimers, isotope selection, and ion mobility need explicit treatment.
• Accurate-mass search uses local TSV databases, not a live HMDB endpoint. The tested 3.6.0 macOS wheel bundles both HMDB mapping and structure tables; inspect your installation and record database versions/checksums. Formula/adduct candidates are putative annotations, not confirmed structures or controlled FDR.
• Alignment failure stops linking unless explicitly overridden with --allow-unaligned. Assess residual RT errors, anchors and missingness. A consensus feature is not necessarily one compound; normalization and missing values require study-specific QC. Isobaric quantification is outside these CLIs.
• GNPS export requires MS2-to-feature annotations with map_index and spectrum_index; a plain MS1 consensus from align_link_quantify.py is insufficient. Exports do not run GNPS/SIRIUS, authenticate, submit data, or validate identities.