Enhancing protein structural properties through model-guided sequence optimization.

Area of Science:

• Biotechnology
• Computational Biology
• Protein Engineering

Background:

• Optimizing proteins for multiple functions (stability, binding, expression) is complex and resource-intensive.
• Traditional protein engineering methods face limitations due to structural complexity and high costs.
• Machine learning (ML) offers a promising avenue for accelerating protein design and optimization.

Purpose of the Study:

• To develop and validate an iterative ML-guided strategy for efficient protein engineering.
• To enhance protein properties such as structural stability, ligand binding affinity, and shape complementarity.
• To reduce the reliance on costly experimental characterization in protein design.

Main Methods:

• An iterative ML-guided approach was employed to explore protein sequence space.
• ML models predicted protein properties, guiding the selection of sequences for experimental validation.
• A genetic algorithm, directed by ML models, identified optimal mutant sequences.
• Experimental data from validated sequences were used to iteratively refine ML models.

Main Results:

• The ML-guided approach successfully identified mutant sequences with improved performance compared to conventional methods.
• Iterative model refinement led to enhanced predictive accuracy with each cycle.
• Novel variants of glutamine binding protein (QBP) with superior stability and binding properties were discovered.
• The method demonstrated efficient exploration of the sequence space for protein optimization.

Conclusions:

• The integration of ML and iterative optimization provides an efficient and scalable solution for protein engineering.
• This approach accelerates the discovery of proteins with tailored functional properties.
• The validated method holds significant potential for advancing various biotechnological applications.