An artificial intelligence-driven synthesis planning platform (PhotoCat) for photocatalysis

Area of Science:

• Synthetic Chemistry
• Computational Chemistry
• Materials Science

Background:

• Photocatalysis is a key technology in modern synthesis.
• AI-driven reaction prediction is limited by data availability.
• Developing efficient photocatalytic processes requires extensive experimental screening.

Purpose of the Study:

• To create a comprehensive database of photocatalytic reactions.
• To develop an AI platform for predicting photocatalytic reactions and conditions.
• To accelerate the discovery of novel photocatalytic transformations.

Main Methods:

• Curated a large-scale, open-source database (PhotoCatDB) of 26.7K photocatalytic reactions with mechanistic details.
• Developed a Transformer-based AI platform (PhotoCat) utilizing 100 million molecular data points.
• Experimentally validated the AI platform's predictions through synthesis.

Main Results:

• PhotoCatDB contains 26.7K reactions, including 9.2K multicomponent reactions.
• The PhotoCat platform achieved high accuracy: 82.6% for reaction prediction, 77.1% for retrosynthesis, and 88.5% for condition recommendation.
• Four novel photocatalytic reactions were discovered and experimentally validated with yields up to 75.3%.

Conclusions:

• The integrated PhotoCatDB and PhotoCat platform represent a significant advancement in AI-assisted photocatalysis.
• This data-driven approach accelerates the discovery of new reactions and optimizes reaction conditions.
• Establishes a new paradigm for sustainable chemistry through computational prediction and experimental validation.