Chapter 13 of Generative

Generative models for molecules soon

Variational autoencoders, SMILES and SELFIES language models, and diffusion and flow models, with the evaluation questions that matter for generated molecules.

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

What it will cover

• Why generate molecules, and how to judge generated sets (validity, uniqueness, novelty)
• Variational autoencoders on fingerprints and SMILES
• Character-level language models for SMILES and SELFIES
• Diffusion and flow models in outline
• Sampling, temperature and mode collapse
• Lab 13: train a small SMILES language model and sample new molecules

Planned learning outcomes

1. Train a generative model of molecules and sample from it.
2. Evaluate generated molecules with validity, uniqueness, novelty and property distributions.
3. Explain the trade-offs between representations (SMILES, SELFIES, graphs) for generation.

In the meantime

Part IV assumes Parts I–III. Start with Chapter 0 if you are new.

PreviousChapter 12 Protein structureNextChapter 14 Generative design loops

Edit this pageReport a problemLast updated 2026-11-02