Protein engineering via Bayesian optimization-guided evolutionary algorithm and robotic experiments.

Area of Science:

• Biochemistry
• Molecular Biology
• Computational Biology

Background:

• Directed protein evolution is crucial for engineering novel protein functions but often limited by experimental throughput.
• Machine learning and robotics have individually shown promise in accelerating protein engineering but haven't been effectively combined.
• Existing methods struggle to leverage the high-throughput capabilities of robotics in conjunction with machine learning for directed evolution.

Purpose of the Study:

• To develop a scalable and batched method for guiding robotic experiments in directed protein evolution.
• To integrate machine learning with robotic platforms for efficient exploration of protein fitness landscapes.
• To enhance the speed and efficiency of protein engineering cycles through a novel algorithm.

Main Methods:

• Development of the Bayesian Optimization-guided EVOlutionary (BO-EVO) algorithm.
• Application of BO-EVO to guide multiple rounds of robotic library creation and screening.
• Testing BO-EVO on empirical landscapes of Protein G domain B1 and Escherichia coli kinase PhoQ, as well as simulated NK landscapes.
• Utilizing BO-EVO for engineering the enzyme specificity of RhlA for rhamnolipid biosurfactants.

Main Results:

• BO-EVO demonstrated successful generalization across different empirical and simulated protein landscapes.
• The algorithm guided robotic experiments to achieve a 4.8-fold improvement in producing a target rhamnolipid congener.
• This significant improvement was achieved by examining less than 1% of all possible mutants over four iterations.
• BO-EVO proved efficient in exploring combinatorial mutagenesis libraries.

Conclusions:

• The BO-EVO algorithm provides an efficient and generalizable approach for guiding combinatorial protein engineering.
• This method effectively integrates machine learning with high-throughput robotic experimentation.
• BO-EVO enables rapid protein engineering with significant improvements in desired traits without requiring prior knowledge.