bio-metabolomics-statistical-analysis

What problem does it solve?

Provides a reproducible, metabolomics-aware statistical pipeline to detect differential metabolites, estimate reliable fold changes, and produce diagnostic visualizations so researchers can move from raw intensity tables to interpretable results without ad-hoc mistakes.

Core Features & Use Cases

• Preprocessing guidance: zero/missing value strategies, log2 transformation, and recommended normalization (PQN, QC-LOESS, VSN) with notes on when normalization can harm interpretation.
• Univariate testing: limma moderated t-tests with eBayes(trend=TRUE, robust=TRUE) for small-n studies, and Welch's t-test / Wilcoxon alternatives for larger or non-normal cases, including BH FDR correction.
• Effect-size handling: clear fold-change computation on log2 data, advice on shrinkage with ashr and minimum-effect testing with treat().
• Multivariate & classification: PCA for QC, PLS-DA / sPLS-DA (mixOmics) and OPLS-DA (ropls) workflows, Random Forest ranking, VIP selection, and ROC/AUC evaluation for biomarker candidates.
• Visualization & outputs: volcano plots, heatmaps, PCA plots, and exportable result tables suitable for pathway analysis and downstream reporting.
• Use case: From an untargeted LC-MS feature table and sample metadata, run preprocessing, choose limma or Python-based testing, generate volcano and PCA plots, and produce a ranked table of candidate metabolites with adjusted p-values and shrunk effect sizes.

Quick Start

Run a full analysis: log2-transform and PQN-normalize my feature table, then run limma with BH correction and return fold changes, adjusted p-values, and a volcano plot.