Designing effective catalysts is a key process for optimizing catalytic reactions, however, existing generative approaches are often limited to specific reaction classes and predefined fragment categories. Here, the authors present CatDRX, a catalyst discovery framework powered by a reaction-conditioned variational autoencoder to generate potential catalysts and predict their activities, integrating optimization and validation based on reaction mechanisms and chemical knowledge.
- Apakorn Kengkanna
- Yuta Kikuchi
- Masahito Ohue
