Machine learning-assisted directed protein evolution with combinatorial libraries.

Area of Science:

• Biochemistry
• Computational Biology
• Protein Engineering

Background:

• Directed protein evolution is crucial for protein engineering but can be experimentally intensive.
• Exploring vast combinatorial sequence space is challenging due to high experimental costs.

Purpose of the Study:

• To integrate machine learning into directed evolution to reduce experimental effort.
• To computationally explore sequence space for identifying improved protein variants.
• To enhance the efficiency and effectiveness of protein engineering.

Main Methods:

• Developed and applied machine learning models trained on experimental data to predict protein variant fitness.
• Validated the approach using a known empirical fitness landscape for a human GB1 binding protein.
• Utilized machine learning-guided evolution to engineer an enzyme for stereodivergent carbene Si-H insertion.

Main Results:

• Machine learning-guided directed evolution identified higher-fitness variants compared to conventional methods.
• The approach successfully evolved an enzyme for stereoselective catalysis, achieving high enantiomeric excess (ee).
• Seven mutations were fixed in two rounds, demonstrating rapid optimization.

Conclusions:

• Machine learning significantly enhances the throughput and quality of directed protein evolution.
• In silico modeling provides a powerful tool for exploring protein sequence space.
• This integrated approach expands the possibilities for protein design and engineering.