Teaching

Use Random Forests to integrate multi-omics datasets, extract proximity measures, and perform downstream analyses including clustering and dimensionality reduction.

Supervised and unsupervised multi-omics integration with mixOmics, including PLS, sPLS, and DIABLO.

Multi-Omics Factor Analysis from its underlying latent-factor model through practical implementation and interpretation.

Kernel-based data integration, custom similarity measures, and combining kernel representations with dimensionality reduction.

Similarity Network Fusion for integrating multiple data types by building sample-similarity networks, exchanging information between them, and producing a fused representation for downstream analysis.

Principal Component Analysis from first principles: geometry, variance maximization, eigenvectors, SVD, visualization, and implementation in R.

Nonlinear dimensionality reduction from pairwise distances and probability distributions through detailed t-SNE and UMAP implementations.

ICA from latent-variable intuition through whitening, non-Gaussianity, FastICA optimization, mathematical derivation, and implementation in R.

NMF from intuition to mathematics, including latent components and scores, practical workflows, interpretation of W and H, and an application to leukemia gene-expression data.

Diffusion maps for nonlinear dimensionality reduction and trajectory-like structure, from local similarity graphs and random walks through diffusion coordinates, practical use in R and Python, mathematical foundations, and exercises.

The theory and practical use of Self-Organizing Maps for high-dimensional data, including training, visualization, clustering, and interpretation.

Density-based clustering from intuition to mathematics, including core, border, and noise points, parameter selection, implementation in R and Python, and exercises using PBMC3k single-cell data.

Spectral clustering from graph construction and similarity matrices through graph Laplacians, eigenvectors, spectral embedding, connected components, and clustering in the embedded space.

Leiden clustering for graph-based community detection, from nearest-neighbour graphs and intuitive community structure through practical use in R and Python, the mathematical optimization procedure, aggregation and refinement, and exercises with PBMC3k single-cell data.

Methods for evaluating and selecting clustering solutions, including internal validation indices, silhouette analysis, gap statistics, stability assessment, consensus clustering, and practical workflows in R and Python.

A conceptual and mathematical introduction to decision trees and Random Forests, including splitting, bagging, feature randomness, and variable importance.

XGBoost from intuition to mathematics: sequential boosting, tree-based error correction, objective functions, regularization, and practical implementation in R.

SHAP for explaining machine-learning predictions, from intuitive local and global interpretation through practical use in R and Python, Shapley values, KernelSHAP, and a worked case study.

LIME for explaining individual machine-learning predictions, from local intuition and neighbourhood generation through practical use in R and Python, locality weighting, surrogate models, mathematical foundations, and exercises with breast-cancer prediction data.

Conformal prediction for quantifying uncertainty in machine-learning predictions, including regression intervals, classification prediction sets, calibration, practical workflows in R and Python, mathematical foundations, and exercises with health survey data.

Linear mixed models from GLS and variance components through ICC, REML versus ML, random effects, and practical modelling with lme4.