Components and scope¶

Component YAML definitions select a registered type and its parameters. Listing or constructing components never downloads resources. The runner calls setup before execution where needed. These tables are generated from the registries.

pride¶

Retrieves selected measurements from a public project. Preparation writes pride.json with catalogue and file metadata. It preserves filenames, native formats and compression. It does not convert RAW to mzML or infer sample metadata. Published checksums are recorded without verifying downloaded contents. Select files explicitly and check their sizes in PRIDE before running.

uniprot¶

Preparation returns the cached response and writes uniprot.json, recording the query and original release headers. Other exports, such as XML, are allowed; consumers decide whether they can read them. Cache reuse does not check for newer releases. The database supplies search candidates, not benchmark ground truth.

sage¶

Quote version strings. Paths resolve relative to the component YAML file. Setup downloads and extracts a pinned Sage release for supported macOS (Intel/ARM), Linux (x86-64/ARM64) or Windows (x86-64) systems. A supplied executable skips downloading. Selected dataset and database paths override inputs in the native JSON. Sage interprets formats and search settings; the wrapper does not validate declared compatibility or convert files.

Each run saves sage-settings.json, the exact input JSON copy Sage executes. sage.command.json records its argument list, working directory and the RAYON_NUM_THREADS override. version.command.json records the version probe. These records are written before execution and remain available on failure. On success, Sage's native results.json records its effective settings, including defaults and input overrides. Native outputs, sage.log and version.txt are preserved. The official 0.14.7 Apple silicon binary reports sage 0.14.6; we preserve both the requested release and executable response. A nonzero exit records a workflow failure. Native warnings and confidence estimates still require interpretation after a successful exit.

comet¶

Setup caches the official Comet binary for macOS/Linux (Intel or ARM64) or Windows (x86-64). Input paths and output locations override the native configuration through Comet's command-line options. Each measurement produces numbered native files, a log and its command record. comet.params preserves the exact settings used. The log includes Comet's version. Gzipped measurements are decompressed once beside their cached source; clearing the shared cache removes both copies. Inputs are never converted to another format.

msamanda¶

Setup caches the official standalone bundle, pinned to both a repository revision and release directory. It includes its .NET runtime. Published builds are Intel x86-64 for macOS, Linux and Windows. Apple Silicon Macs require Rosetta, installed separately by the user; Linux ARM requires a supplied executable. The macOS download is about 46.3 MB. MS Amanda is distributed under its own freeware license.

Each run saves the original msamanda-input.xml, the resolved msamanda-settings.xml, and copies of the enzyme/modification definitions. Definitions explicitly named in the input XML resolve relative to that file; otherwise the bundled definitions are used. Runtime data stays under the run's runtime/ directory. Numbered native CSV files, logs and command records preserve each measurement's output; the native banner records the executable version. Gzip preparation is shared with Comet. No rescoring tool is installed or run.

Expected peptides¶

expected_peptides reads Sage's results.sage.tsv, Comet's numbered .comet.txt files or MS Amanda's numbered .msamanda.csv files. It selects rank-one target matches and counts each peptide once across measurements. Comet's default DECOY_ and Amanda's REV_ protein prefixes identify decoys. Keep these prefixes when configuring native decoy generation.

Sage reports recovery before and after filtering on spectrum_q; exact Sage modification notation is required unless sequence_only: true is selected. Comet and Amanda require sequence_only: true and max_q_value: null: their native scores are not q-values. Reports explicitly label this recovery as unfiltered and omit confidence-threshold metrics. An empty, correctly headed table gives zero recovery; missing columns or invalid target rows raise an error.

expected_peptides.json records the reference, matching mode, confidence filter, matching/missing peptides, source tables and metrics. Expectations must come from an independent reference. This diagnostic does not estimate precision, FDR or quantification accuracy, and unlisted peptides are not assumed incorrect. No extras are needed.

ProteoBench DDA¶

Install with poetry install --extras proteobench. proteobench_dda uses ProteoBench 0.18.6 to score Sage's lfq.tsv and results.json for the published Q Exactive DDA HYE design. It requires all six runs from PXD028735:

LFQ_Orbitrap_DDA_Condition_A_Sample_Alpha_01.mzML
LFQ_Orbitrap_DDA_Condition_A_Sample_Alpha_02.mzML
LFQ_Orbitrap_DDA_Condition_A_Sample_Alpha_03.mzML
LFQ_Orbitrap_DDA_Condition_B_Sample_Alpha_01.mzML
LFQ_Orbitrap_DDA_Condition_B_Sample_Alpha_02.mzML
LFQ_Orbitrap_DDA_Condition_B_Sample_Alpha_03.mzML

.mzML.gz is also accepted. Preserve _HUMAN, _YEAST and _ECOLI protein flags; expected A/B intensity ratios are 1, 2 and 0.25. Enable Sage LFQ while keeping charge states separate in its native configuration:

{"quant": {"lfq": true, "lfq_settings": {"combine_charge_states": false}}}

The adapter does not add confidence filtering or impute missing intensities. Reports contain metrics and provenance (proteobench.json), a per-ion table (ions.tsv) and copies of packaged settings. Metrics cover quantified ion count, ratio error and within-condition variation; undefined metrics are named separately. Evaluation is local, with no leaderboard submission or reference download.

This adapter cannot score the first experiment's MGF peptide-identification output. ProteoBench also offers a small offline walkthrough using bundled precomputed MaxQuant output. That exercises its scorer on a laptop, but does not rerun Sage on a smaller version of the six-run reference experiment.