Chapter 6 of Cheminformatics

QSAR and property prediction soon

Quantitative structure–activity relationships done honestly: scaffold splits, applicability domain, uncertainty, and the leakage traps of chemical data.

soon This chapter is planned and not yet written. The outline below is the current plan; it may change.

What it will cover

• The QSAR workflow from assay data to model
• Scaffold splits and time splits versus random splits
• Applicability domain: when not to trust a prediction
• Uncertainty estimates with ensembles and conformal methods
• Activity cliffs and near-duplicates as leakage
• Lab 6: a scaffold-split benchmark on a MoleculeNet dataset

Planned learning outcomes

1. Build a QSAR model with a scaffold split and explain why its score differs from a random split.
2. Estimate an applicability domain and flag out-of-domain predictions.
3. Attach calibrated uncertainty to property predictions and communicate it to a chemist.

In the meantime

Everything here builds on Part I, in particular the evaluation habits of Chapter 2. Episode 2.6 already introduces the group splits that become scaffold splits here.

PreviousChapter 5 Molecules as dataNextChapter 7 Graph neural networks for molecules

Edit this pageReport a problemLast updated 2026-11-02