An explainability framework for deep learning on chemical reactions exemplified by enzyme-catalysed reaction...

Area of Science:

• Biochemistry
• Computational Chemistry
• Biocatalysis

Background:

• Enzyme-catalyzed reactions are crucial in life sciences and chemistry.
• Current enzyme assignment relies on expert knowledge due to limited data.
• There's a need for automated methods to associate enzymes with reactions.

Purpose of the Study:

• To develop a data-driven machine learning approach for enzyme-reaction association.
• To support and automate the identification of catalytic enzymes for biochemical reactions.
• To predict candidate enzymes for biocatalysis in organic synthesis.

Main Methods:

• A human-in-the-loop machine learning strategy was employed.
• The approach utilizes data-driven methods for enzyme-reaction mapping.
• Explainability and visualization tools were integrated into the ML model.

Main Results:

• The method successfully associates enzymes with biochemical reactions.
• It predicts potential biocatalysts for organic reactions.
• Explainable AI (XAI) features enhance understanding of the predictions.

Conclusions:

• The developed machine learning approach automates enzyme assignment for biochemical reactions.
• It offers a powerful tool for metabolic engineering and sustainable chemistry.
• The explainability framework is adaptable to other chemical and biochemical ML applications.