An integrative approach to protein sequence design through multiobjective optimization

Area of Science:

• Computational biology
• Protein engineering
• Bioinformatics

Background:

• Deep learning advances necessitate integrated frameworks for computational protein design.
• Current methods struggle to coherently combine diverse models and objectives in generative design.

Purpose of the Study:

• To adapt evolutionary multiobjective optimization for integrating multiple models and objectives in protein design.
• To improve the efficiency and accuracy of generative protein design processes.

Main Methods:

• Utilized Non-dominated Sorting Genetic Algorithm II (NSGA-II) as the core optimization framework.
• Integrated AlphaFold2 and ProteinMPNN confidence metrics for objective space definition.
• Developed a mutation operator using ESM-1v and ProteinMPNN for sequence redesign.

Main Results:

• Demonstrated significant reduction in bias and variance for RfaH native sequence recovery compared to direct ProteinMPNN application.
• Showcased the approach's effectiveness on fold-switching protein RfaH and higher-dimensional problems (PapD, calmodulin).
• Observed improvements attributed to an informative mutation operator, parallel iterative design, and Pareto front approximation.

Conclusions:

• Evolutionary multiobjective optimization offers a robust framework for complex protein design tasks.
• The proposed method enhances sequence space exploration and design candidate diversity.
• This adaptable approach is broadly relevant for generative protein design with intricate specifications.