orangemetabo — MetaboAnalyst-Style Statistics for Orange3

Add-on for Orange3 that reproduces the MetaboAnalyst univariate workflow for GC-MS / GC-IMS feature tables (Compound Discoverer exports), as used in the coffee-fermentation study (AK Weller).

Pipeline

Feature Table CSV ──> Preprocess ──> [Filter] ──> Univariate Stats ──> Heatmap
 (2-header, ;)       (Sum-Norm,       (QC)         (ANOVA / Welch /     (Top-N,
                       log2, Imputation,             Kruskal + BH-FDR     Ward cluster,
                       Autoscale/Pareto)             + log2FC + means)     PNG/SVG export)
                                                                           └─> Volcano
                                                                              (log2FC vs -log10 FDR)

Widgets (category Metabo Weller)

Correctness

orangemetabo/metabo_core.py is the headless, Qt-free analytics core. The default pipeline (sum-normalise → log2 → autoscale → one-way ANOVA → BH-FDR) reproduces the validated ground truth analysis_CV/cv_anova_alle_97_features.csv exactly: 97/97 features on F, p, and FDR (max abs diff < 5e-5). Reproduce with _test_core.py (core) and _test_e2e.py (all six widgets, offscreen) — both run against the Dropbox ground truth and print 97/97. The volcano contrast is checked to equal the Welch log2FC and the ground-truth log2FC_ANF_vs_WILD column.

Note on the volcano's fold change: it is the difference of the per-group means in the space of the data you feed in. For a real log2 fold change, preprocess with Sum normalisation + log2 and no scaling (Metabo Preprocess → Method "Sum (total area)", log2 on, Scaling "None"). With autoscaled/Pareto data the x-axis is a scaled mean difference (the p-values are unaffected, since the t-test is scale-invariant per feature).

Installation

cd orange-metabo-addon
/Applications/Orange.app/Contents/Frameworks/Python.framework/Versions/Current/bin/python3 -m pip install -e .

Or from the monorepo: python ../orange-install.py metabo.