Skip to content

notame

Workflow for non-targeted LC-MS metabolic profiling — a Bioconductor package covering tabular preprocessing, quality control, and uni-/multi-variate analysis of untargeted LC-MS metabolomics data, packaging the recommendations from the Metabolomics Data Processing and Data Analysis — Current Best Practices special issue.

Talk at EuroBioC2025

  • Speaker: Vilhelm Suksi (University of Turku)
  • Day / session: Day 1 (Wed Sep 17, 2025) — morning short-talk session, after the Crowell keynote
  • Talk title: "notame package for LC–MS metabolomics analysis"
  • Slides (Zenodo DOI): to verify after publish
  • Video: not recorded
  • Abstract / schedule: eurobioc2025 schedule

What it does

notame is a one-stop workflow for untargeted LC-MS metabolomics: it ingests feature tables, runs QC visualizations and drift correction, performs univariate and multivariate analysis (PCA, PLS-DA, etc.), and emits diagnostic reports — a curated translation of the field's best-practice recommendations into runnable R.

How it works

Core idea. notame layers QC-pooled-sample-aware preprocessing on top of a SummarizedExperiment: it models analytical drift per feature with cubic-spline regression fit through the pooled QC injections, then flags features by detection rate and robust dispersion metrics (RSD*, D-ratio*) computed on those same QCs. Flagged features are not deleted — they are kept in the object but silently excluded from downstream operations (PCA, imputation, statistics) so the user can audit them.

Inputs / outputs. Input: a feature × sample abundance table plus sample metadata identifying pooled QCs and injection order, wrapped in a SummarizedExperiment. Output: the same object with drift-corrected assays, per-feature QC flags in rowData, missing values imputed (random-forest), and figures / Rmd-style diagnostic reports.

Key innovation. The flagging-not-filtering convention plus codifying the Metabolomics Data Processing — Current Best Practices recommendations into a single importable pipeline. Drift correction is the cubic-spline-on-QCs approach standard in the field rather than a novel method.

Parameters worth knowing. Spline smoothing parameter range (default 0.5–1.5, CV-selected). Detection-rate threshold (default ≥0.5). Robust RSD* limit (≤0.2 recommended). Robust D-ratio* limit (≤0.4 recommended).

Canonical example. Vignette uses the bundled toy_notame_set: call mark_nas() to convert zeros to NA, correct_drift() for spline-based drift correction, flag_detection() / flag_quality() to mark poor features against the QC thresholds, then impute_rf() for random-forest imputation. The result is a QC-cleaned SummarizedExperiment ready for PCA / PLS-DA.

Where it fits in the corpus

  • AACR 2026: no current dossier; relevant axis is bioinfo / AI methods (metabolomics is adjacent to the cancer-metabolism talks)
  • GBCC 2025: queued
  • Nextflow Summit 2026: not mentioned

Notes

Mature workflow package (Bioconductor 3.23, v1.2.0). The value is curation more than novelty — converting consensus best-practice into one importable pipeline.