Adaptive machine learning for protein engineering.

Area of Science:

• Biochemistry
• Computational Biology
• Machine Learning

Background:

• Machine-learning models are increasingly used to predict protein function from sequence.
• Protein engineering faces challenges due to the vast number of possible protein sequences.
• Sequence-to-function models offer a promising approach to guide protein design.

Purpose of the Study:

• To review methods for using machine-learning surrogate models to select protein sequences for experimental measurement.
• To outline strategies for optimizing protein sequence selection in protein engineering.

Main Methods:

• Discusses single-round machine-learning optimization for sequence selection.
• Explains sequential optimization for iterative model improvement and sequence discovery.
• Highlights the use of machine-learning surrogate models in protein engineering workflows.

Main Results:

• Machine-learning models can effectively guide the selection of protein sequences with desired functions.
• Both single-round and sequential optimization strategies are viable for discovering novel protein sequences.
• Sequential optimization allows for continuous model refinement and enhanced discovery of optimized proteins.

Conclusions:

• Machine-learning models are powerful tools for navigating the complexity of protein sequence space.
• Effective strategies for sequence selection are crucial for successful protein engineering.
• Iterative optimization enhances both the discovery of functional proteins and the predictive accuracy of the models.