Machine learning-assisted enzyme engineering.

Area of Science:

• Protein Engineering
• Computational Biology
• Biotechnology

Background:

• Traditional enzyme engineering methods like directed evolution and rational design face challenges due to vast protein sequence space and combinatorial complexity.
• Experimental approaches are often limited, especially when protein structure-function relationships are unknown, hindering the full exploration of natural enzyme potential.

Purpose of the Study:

• To review recent advancements in machine learning (ML)-assisted enzyme engineering.
• To highlight ML's role in overcoming limitations of traditional enzyme engineering strategies.
• To discuss strategies combining ML with experimental methods for effective enzyme reengineering.

Main Methods:

• Utilizing advancements in next-generation sequencing, high-throughput screening, protein databases, and artificial intelligence (AI), particularly ML.
• Applying ML to accurately determine quantitative structure-property/activity relationships for enzyme property prediction.
• Combining experimental library generation with ML-based prediction for accelerated enzyme engineering.

Main Results:

• ML-assisted approaches have demonstrated efficiency and accuracy in predicting diverse enzyme properties.
• Successful examples of ML application include enhancing enzyme activity, enantioselectivity, and thermostability.
• ML integration accelerates the enzyme engineering cycle compared to traditional methods.

Conclusions:

• Data-driven enzyme engineering using ML offers a promising solution to challenges in protein engineering.
• Combining random or (semi-)rational approaches with ML enables effective enzyme reengineering for improved targeted properties.
• Addressing current limitations and future opportunities is crucial for realizing the full potential of ML in enzyme engineering.