proteomics-ms-qc

proteomics-ms-qc

When to use

The user has a protein-quantification CSV (typically the output of proteomics-data-import, with rows = proteins and columns = samples + metadata) and wants QC summary statistics: protein count, sample count, fraction of missing intensities, per-protein coefficient of variation (CV) — median and mean. Auto-detects intensity columns by select_dtypes(include=[np.number]).

This skill does NOT process raw spectra. For peptide / PSM-level identification stats use proteomics-identification.

Inputs & Outputs

Inputs

• Modalities: ms
• File types: .csv

Outputs

• tables/qc_metrics.csv
• report.md
• result.json

Flow

1. Load CSV (--input <file.csv>) or generate a demo at output_dir/demo_proteomics.csv (proteomics_ms_qc.py:223).
2. Detect numeric (intensity) columns via select_dtypes(include=[np.number]) (proteomics_ms_qc.py:47); raise ValueError("No intensity/sample columns detected in input data") at :74 if none found.
3. Compute n_proteins / n_samples / missing_rate / per-protein CV.
4. Write tables/qc_metrics.csv (proteomics_ms_qc.py:241) + report.md + result.json.

Gotchas

• Sample columns must be NUMERIC. Intensity-column auto-detection (proteomics_ms_qc.py:47) uses select_dtypes(include=[np.number]). String-typed intensities (e.g. quoted numbers in some Spectronaut exports) are silently treated as metadata, not samples — your n_samples will be 0 and the run raises ValueError at :74.
• No intensity columns ⇒ hard fail. proteomics_ms_qc.py:74 raises ValueError("No intensity/sample columns detected in input data") — there is no auto-detection of intensity_* prefixes; only dtype-based.
• --input REQUIRED unless --demo. proteomics_ms_qc.py:228 raises ValueError("--input required when not using --demo").
• Both NaN and 0.0 count as missing. proteomics_ms_qc.py:80 computes missing_mask = np.isnan(intensities) | (intensities == 0) — zero is treated as "not detected" (the proteomics convention). If your search engine writes a small placeholder (e.g. 1.0) for undetected proteins, the missing rate is artificially LOW; pre-impute placeholders to 0 or NaN first.
• CV is per-protein across samples. Reported median_cv / mean_cv are aggregations across the per-protein CV distribution — interpret as "typical protein-level reproducibility", not "sample-level reproducibility".