PhotoCat: An Artificial Intelligence-Driven Synthesis Planning Platform for Photocatalysis DOI Creative Commons
Jiangcheng Xu, An Su,

Panyi Huang

et al.

Published: Sept. 12, 2023

Photocatalysis is becoming increasingly important in modern chemistry for efficient multicomponent one-pot synthesis. However, predicting the results of photocatalytic reactions using artificial intelligence remains challenging, mostly due to insufficient number and incomplete information on reaction conditions existing databases. In this study, we curated Database (PhotoCatDB), which consists 6,523 (of 6,175 are multicomponent) containing condition such as photocatalysts, bases or acids, additives, solvents. Before adding training data, attention-based deep learning model PhotoCat pre-trained USPTO fine-tuned PhotoCatDB had a Top-1 accuracy 78.16%, was 77.70% higher than same trained only database 14.53% by from Reaxys. After further increased 82.25%. addition, interpretability reflected its attention weights, can infer model’s understanding chemistry. Furthermore, five previously unreported predicted were successfully validated wet-lab experiments, demonstrating potential identifying verifying novel photocatalysis real-world significance.

Language: Английский

PhotoCat: An Artificial Intelligence-Driven Synthesis Planning Platform for Photocatalysis DOI Creative Commons
Jiangcheng Xu, An Su,

Panyi Huang

et al.

Published: Sept. 12, 2023

Photocatalysis is becoming increasingly important in modern chemistry for efficient multicomponent one-pot synthesis. However, predicting the results of photocatalytic reactions using artificial intelligence remains challenging, mostly due to insufficient number and incomplete information on reaction conditions existing databases. In this study, we curated Database (PhotoCatDB), which consists 6,523 (of 6,175 are multicomponent) containing condition such as photocatalysts, bases or acids, additives, solvents. Before adding training data, attention-based deep learning model PhotoCat pre-trained USPTO fine-tuned PhotoCatDB had a Top-1 accuracy 78.16%, was 77.70% higher than same trained only database 14.53% by from Reaxys. After further increased 82.25%. addition, interpretability reflected its attention weights, can infer model’s understanding chemistry. Furthermore, five previously unreported predicted were successfully validated wet-lab experiments, demonstrating potential identifying verifying novel photocatalysis real-world significance.

Language: Английский

Citations

1