Reaction-conditioned generative model for catalyst design and optimization with CatDRX

Area of Science:

• Catalysis
• Computational Chemistry
• Materials Science

Background:

• Catalyst design is crucial for optimizing chemical reactions, reducing waste, and improving efficiency.
• Current generative models for catalyst discovery are often limited to specific reaction types and predefined molecular fragments.
• A broader approach is needed to explore novel catalysts across diverse reaction spaces.

Purpose of the Study:

• To present CatDRX, a catalyst discovery framework utilizing a reaction-conditioned variational autoencoder.
• To generate novel catalysts and predict their performance for various chemical reactions.
• To advance catalyst design and discovery in the chemical and pharmaceutical industries.

Main Methods:

• Developed a reaction-conditioned variational autoencoder generative model (CatDRX).
• Pre-trained the model on a large reaction database and fine-tuned it for specific downstream reactions.
• Integrated property optimization and validation based on reaction mechanisms and chemical knowledge.

Main Results:

• Achieved competitive performance in predicting reaction yield and catalytic activity.
• Successfully generated potential catalysts tailored to specific reaction conditions.
• Demonstrated the framework's effectiveness through various case studies.

Conclusions:

• CatDRX offers a versatile approach to catalyst design and discovery.
• The framework facilitates the identification of novel catalysts by considering reaction components.
• This work advances catalyst development for industrial applications, including pharmaceuticals.