log-RRIM: Yield Prediction via Local-to-Global Reaction Representation Learning and Interaction Modeling.

Area of Science:

• Organic Chemistry
• Computational Chemistry
• Artificial Intelligence

Background:

• Accurate chemical reaction yield prediction is vital for efficient organic synthesis.
• Artificial intelligence (AI) offers potential for accelerating yield predictions.
• Current methods may not fully capture complex reactant-reagent interactions.

Purpose of the Study:

• To introduce log-RRIM, a novel graph transformer-based framework for predicting chemical reaction yields.
• To leverage AI for in silico prediction of reaction outcomes, reducing experimental efforts.
• To enhance the accuracy of yield prediction by modeling key chemical principles.

Main Methods:

• Developed log-RRIM, a graph transformer framework incorporating a cross-attention mechanism.
• Implemented a local-to-global reaction representation learning strategy.
• Trained and evaluated the framework on predicting chemical reaction yields.

Main Results:

• log-RRIM demonstrated superior performance in predicting reaction yields.
• The framework showed particular strength in predicting medium-to-high-yielding reactions.
• The cross-attention mechanism effectively modeled reactant-reagent interplay.

Conclusions:

• log-RRIM is a reliable AI tool for chemical reaction yield prediction.
• The framework's ability to model interactions and contributions enhances reaction planning.
• This approach offers a valuable method for optimizing organic synthesis.