Skip to content

← Teaching

PhD programme in Chemical and Molecular Sciences · 2026/27

Chemometrics for Mass Spectrometry-Based Metabolomics

Metabolomics data are not a list of metabolites: they are a matrix of signals, shaped by every choice made before any statistics. This course follows that matrix from the sample to the interpretation, and teaches how to use chemometric methods without fooling yourself:

  • how to design a study;
  • how to build and clean the data matrix;
  • how to explore it;
  • how to build and validate a predictive model;
  • how to test thousands of variables at once;
  • how far the name given to a signal can be trusted.

Every method is applied to real, public data, and every result is put to the test with labels assigned at random, to see how much an analysis “discovers” when there is nothing to discover. Sessions are active: predictions to check, calculations in pairs, and critical reading of methods sections from published papers.

Programme

8 sessions of 2 hours.

  1. From sample to data matrix: study design, MS acquisition, quality control
  2. LC–MS data preprocessing: from raw files to a matrix ready for statistics
  3. Unsupervised analysis: PCA, outliers, clustering, t-SNE and UMAP maps
  4. PLS-DA and validation: cross-validation, leakage, permutation tests
  5. Univariate statistics and multiple testing: effect sizes, FDR, unit of analysis, power
  6. Machine learning: penalised regression, SVMs, trees, forests, fair comparison of models
  7. Identification of metabolites and lipids: confidence levels, exact mass, isotopes, MS/MS libraries
  8. Case study: re-analysis of a published plasma lipidomics and metabolomics study, starting from the files shared by its authors

Materials. For each session, the lecture slides and a reference text with worked examples and review questions with solutions.

Data. Public datasets from MetaboLights (MTBLS79, MTBLS2, MTBLS404), MassBank spectra and a clinical study shared on MassIVE.

Who it is for. PhD students who produce or analyse metabolomics and lipidomics data. Prerequisites:

  • molecular formulas and isotopes at the level of general chemistry;
  • mean, standard deviation and correlation.

Official title in the PhD programme: “Metabolomic approaches in mass spectrometry: chemometric and data-driven analysis”.

People