Machine Learning in Bioinformatics

Learning from patterns in molecular biology data plays an important role in diagnosing disease, discovering new targets for therapy, and more generally in answering biological questions that lead to an improved understanding of biological systems with relevance to human health, industry, biotechnology, and agriculture.

This course focuses on methodology for the analysis of high-dimensional data in molecular biology, naturally addressing challenges that arise in the field such as learning from unlabelled data or from small numbers of samples. The methodology is introduced in the context of real-world applications with examples using real data.

Covered topics will include (a selection of) the following:

• Mixture models: (in)finite mixture models, EM algorithm, ECDF, bootstrapping.

• Clustering: k-means/medoids, density-based clustering, hierarchical clustering, clustering evaluation, choosing number of clusters.

• Statistical hypothesis testing: p-values, confidence intervals, multiple hypothesis testing, family-wise error rate, (local) false discovery rate.

• Testing using high-throughput count data: multifactorial designs, (generalized) linear models, analysis of variance, shrinkage estimation.

• Linear and non-linear dimensionality reduction: matrix decomposition (SVD), biplot representations, PCA, NMF, UMAP, autoencoders.

• Multivariate methods for heterogenous data: embeddings, multidimensional scaling (MDS), robust (non-metric) MDS, correspondence analysis, canonical correlation.

• Supervised learning: algorithms, performance metrics, curse of dimensionality, generalizability and model complexity, regularization, cross-validation, ML architectures for sequential data.

• Interpretability and explainability.

• Other selected advanced topics, including transfer learning self-supervised learning, semi-supervised learning, multimodal ML.