Reproduce “Predicting Chemical Reaction Outcomes Based on Electron Movements Using Machine Learning” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Shuan Chen, Kye Sung Park, Taewan Kim, Sunkyu Han, Yousung Jung

Researchers from Seoul National University and KAIST developed Reactron, the first machine learning model to predict chemical reaction outcomes by explicitly simulating electron movements. This approach achieved 96.4% product accuracy and 94.8% mechanism accuracy on the USPTO dataset, while also demonstrating few-shot learning for complex reactions and enabling the discovery of novel chemical transformations.

Key insights

  • Focusing on explicit electron flow as the fundamental unit of prediction allows the model to simulate human-like mechanistic reasoning.
  • A reactive pooling mechanism efficiently identifies and filters potential reaction centers, significantly reducing computational complexity.
  • A Transformer-like attention mechanism enhances selectivity prediction for potential electron pathways by enabling information exchange between candidates.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.