AI for Drug Discovery
Learn how machine learning is used to find drug targets, screen compounds and predict ADMET properties — with hands-on work on real protein–ligand data.
Course detailsThe core discipline
Biological datasets are small, correlated and batch-affected, which is exactly the setting where machine learning most easily lies. The centre of this course is therefore not model architecture but evaluation: how the split is built, what the metric rewards, and whether the result would survive a new batch of samples.
You will be asked, more than once, to show that a strong result is not an artefact — and to report it as an artefact when it is.
Assessment
Each fortnight ends with an applied analysis on a public dataset: expression, imaging or sequence data, depending on the track you choose. The capstone is a short report in which you must beat a baseline you defined yourself, and document what your model fails to do.
Syllabus
Weeks 1–2
From question to target
What a model can and cannot answer, label design, class imbalance, and evaluation metrics that match the decision being made.
Weeks 3–4
Data hygiene
Leakage, batch effects, duplicates and near-duplicates, and building a split that respects biological structure.
Weeks 5–6
Models for biological data
Classical baselines, regularisation, tree ensembles, and deep models for sequences and images — with a baseline you must beat.
Weeks 7–8
Interpretation and honesty
Feature attribution, calibration, applicability domain, and writing an analysis that states its own limits.
01/Keep going
Python for Bioinformatics
Go from your first script to a working sequence-analysis pipeline in Python — reading FASTA and FASTQ, working with Biopython, and reproducing a result end to end.
Course detailsResearch hub
Method explainers that pair with this course: how the calculations work, and where their assumptions stop holding.
Read the hub