Cheminformatics

• QSAR modelling and applicability domain estimation
• ADMET prediction and the endpoints where models remain unreliable
• Scaffold splitting instead of random splitting for honest evaluation

Genomics

• High-dimension low-sample-size problems and regularisation
• Batch effects and correction methods that can destroy real signal
• Multi-omic integration and interpretability of the resulting features

Tooling

• RDKit, scikit-learn and Bioconductor in a reproducible pipeline
• Experiment tracking and environment capture for a wet-lab collaboration
• Version control and data management for a research group

Validation

• External validation and the reproducibility crisis in biomedical ML
• Prospective testing against retrospective performance
• Reporting standards and reviewer expectations for ML in biomedical journals