EquiReact: An Equivariant Neural Network for Chemical Reactions

Explore the cutting-edge developments in chemical reaction property prediction through this comprehensive talk on EquiReact, an innovative equivariant neural network. Delve into the challenges of reaction property prediction and discover how EquiReact utilizes three-dimensional structures of reactants and products to infer reaction properties. Learn about the model's competitive performance in predicting activation barriers across various datasets and its advantages over existing state-of-the-art models. Gain insights into EquiReact's flexibility, extrapolation capabilities, and data efficiency. Understand the interpretation of latent representations and how EquiReact leverages different information compared to chemically-structured baselines. Follow along as the speaker covers the background, introduces 3DReact, discusses relevant datasets, presents results, and concludes with a Q&A session.

- Intro + Background
- 3DReact
- Datasets
- Results
- Conclusions
- Q&A