First experiment¶
This laptop experiment connects PRIDE + UniProt → Sage → peptide evaluation. It searches one small processed peak list from PXD000001, an Erwinia sample with four protein spikes. Sage reads the compressed MGF directly; no RAW conversion or separate Sage installation is needed.
From the repository root, with Python 3.12+ and Poetry installed:
poetry install
poetry run interop-trace run erwinia_sage --output results/simple-001
Use a new output directory each time. If results/simple-001 already exists,
choose another name. Add --check to validate configuration without downloads or
execution. Every experiment requires an evaluator.
Inputs and download sizes¶
Sizes were checked on 18 October 2026, before downloading the measurement.
| Input | Download |
|---|---|
PRIDE_Exp_Complete_Ac_22134.pride.mgf.gz |
5,984,662 bytes (5.71 MiB), verified by HTTP HEAD and actual file size |
| UniProt FASTA | 2,138,744 bytes (2.04 MiB) in the verified run |
| Sage 0.14.7 | About 6.6 MB on Apple silicon, only if not already cached |
The PRIDE catalogue reports an outdated 16,448,103-byte measurement size; the download server reports the actual compressed size above. This is the only selected measurement. The RAW, mzXML and full project are not downloaded. These sizes document the verified files; the provider does not enforce a download-size limit.
The database query includes the Erwinia strain SCRI 1043 (taxonomy 218491) and spikes P00924, P02769, P00489 and P62894. The verified UniProt response contains 4,467 entries, including the background proteome rather than just expected positives. Downloads reuse the existing PRIDE provider and shared cache. UniProt is a live query: a later fresh download can differ; its release is recorded in each run.
Search and evaluation¶
Native Sage settings follow the deposited Mascot settings: trypsin with two missed cleavages, ±25 ppm precursor and ±0.2 Da fragment tolerances, fixed TMT6plex labels on N-termini/lysine and methylthio on cysteine, plus variable methionine oxidation and TMT6plex on tyrosine. Retention-time prediction is disabled.
The deposited reference results provide four selected reference peptides, one per spike:
| Spike | Reference peptide |
|---|---|
| Yeast enolase, P00924 | GNPTVEVELTTEK |
| Bovine albumin, P02769 | HLVDEPQNLIK |
| Rabbit phosphorylase, P00489 | EIWGVEPSR |
| Bovine cytochrome c, P62894 | TGPNLHGLFGR |
These were chosen before running Sage. The benchmark configuration includes their
fixed TMT labels in Sage's exact sequence notation. Evaluation counts their recovery
among rank-one target matches before and after spectrum_q <= 0.01.
The verified laptop run recovered all four, including at that threshold.
This is consistency with selected deposited identifications, not independent proof of accuracy or FDR calibration. The processed peak list and selected reference peptides do not represent an unbiased full-RAW benchmark. This experiment does not score the project's TMT reporter ratios or ProteoBench LFQ quantification.
Where results go¶
Under the selected output directory:
| File | Evidence |
|---|---|
configuration.json |
Original YAML, component choices and package version |
erwinia_sage/dataset/pride.json |
PRIDE catalogue, selected source and download record |
erwinia_sage/database/uniprot.json |
Query, release and retrieval time |
erwinia_sage/workflow/sage-settings.json |
Exact input settings copy executed by Sage |
erwinia_sage/workflow/sage.command.json |
Command arguments, working directory and thread override |
erwinia_sage/workflow/results.json |
Sage's effective settings, including defaults and input overrides |
erwinia_sage/workflow/version.txt |
Version reported by the actual executable |
erwinia_sage/workflow/sage.log |
Native log and warnings |
erwinia_sage/workflow/results.sage.tsv |
Native peptide-spectrum matches |
erwinia_sage/evaluation/expected_peptides.json |
Reference recovery and threshold |
erwinia_sage/experiment.json |
Execution status, artifact paths and metrics |
New runs save the input snapshot and command records; older runs retain only the files recorded at the time. Inputs remain in the shared cache rather than being duplicated under results. The run is therefore not a self-contained archive: clearing the cache removes its inputs. Locally computed content hashes, the code revision and a complete dependency/environment snapshot are not currently recorded.
See configuration for other combinations and components for parameter scope.