Active learning-assisted directed evolution

Area of Science:

• Biochemistry
• Protein Engineering
• Computational Biology

Background:

• Directed evolution (DE) is crucial for optimizing protein function.
• DE struggles with complex mutation interactions (epistasis).
• Efficient protein optimization requires advanced strategies.

Purpose of the Study:

• Introduce Active Learning-assisted Directed Evolution (ALDE) as an improved DE workflow.
• Address the challenge of optimizing proteins with epistatic mutations.
• Enhance the efficiency of protein engineering.

Main Methods:

• Developed an iterative machine learning-assisted DE workflow (ALDE).
• Utilized uncertainty quantification for efficient search space exploration.
• Applied ALDE to optimize five epistatic residues in an enzyme's active site.

Main Results:

• Achieved a significant yield improvement from 12% to 93% for a non-native cyclopropanation reaction in three experimental rounds.
• Computational simulations on existing datasets support ALDE's superiority over traditional DE.
• Demonstrated ALDE's effectiveness in a challenging protein engineering scenario.

Conclusions:

• ALDE offers a practical and broadly applicable strategy for protein engineering.
• ALDE overcomes limitations of traditional DE, especially with epistatic mutations.
• This approach unlocks improved protein engineering outcomes and expands enzyme applications.