HiRXN: Hierarchical Attention-Based Representation Learning for Chemical Reaction.

Area of Science:

• Organic Chemistry
• Computational Chemistry
• Cheminformatics

Background:

• Natural Language Processing (NLP) and Large Language Models (LLMs) are increasingly used in organic chemistry.
• Effective chemical reaction representation is crucial for NLP models in chemistry prediction.
• Existing methods often overlook the hierarchical and structural information in chemical reactions.

Purpose of the Study:

• To develop a novel tool, HiRXN, for comprehensive chemical reaction representation.
• To capture the hierarchical structure of chemical reactions for improved machine learning.
• To enhance feature engineering for machine learning models in organic chemistry.

Main Methods:

• HiRXN utilizes a novel tokenization method, RXNTokenizer, to capture atomic microenvironment features.
• A hierarchical attention network integrates atomic and molecular level information.
• The approach focuses on learning representations based on the inherent hierarchical structure of reactions.

Main Results:

• HiRXN demonstrates capability in representing chemical reactions effectively.
• The tool achieves remarkable performance in reaction regression and classification tasks.
• Experimental results validate the enhanced feature engineering and representation learning.

Conclusions:

• HiRXN provides a superior method for chemical reaction representation by considering hierarchical structure.
• The developed tool improves the accuracy of machine learning-based chemistry prediction tasks.
• A web server is available for practical application of HiRXN, accepting Reaction SMILES input.