From machine learning to multimodal models: The AI revolution in enzyme engineering

Area of Science:

• Biochemistry and Molecular Biology
• Computational Biology and Bioinformatics
• Synthetic Biology

Background:

• Enzyme engineering is crucial for synthetic biology, biocatalysis, and drug discovery.
• Recent AI advancements, including machine learning and protein models, are driving a data-driven era in enzyme engineering.
• A comprehensive understanding of AI's role and evolution in this field is essential.

Purpose of the Study:

• To provide a guidance map of current enzyme engineering tasks.
• To offer an integrative perspective on AI methods, model types, landmark tasks, and data resources.
• To outline the progression of AI in enzyme engineering.

Main Methods:

• Delineating core modeling tasks: function annotation, structural modeling, and property prediction.
• Reviewing recent advances and dominant algorithmic frameworks in AI for enzyme engineering.
• Tracing the evolution of AI through four stages: classical ML, deep neural networks, protein language models (pLMs), and multimodal architectures.

Main Results:

• Identified key AI methods and model types applied to enzyme engineering tasks.
• Detailed the historical progression of AI techniques in the field.
• Highlighted four transformative trends: unified embeddings, multimodal systems, intelligent agents, and dynamic function simulation.

Conclusions:

• AI is rapidly advancing enzyme engineering, moving towards intelligent, generalizable, and interpretable platforms.
• The integration of AI facilitates data-driven enzyme design for diverse applications.
• Future directions point towards more sophisticated AI systems capable of complex reasoning and dynamic function simulation.