An integrative approach to protein sequence design through multiobjective optimization.

Area of Science:

• Computational protein design
• Bioinformatics
• Structural biology

Background:

• Deep learning advances necessitate integrated frameworks for generative protein design.
• Current methods face challenges in coherently combining multiple models and objective functions.

Approach:

• Adapted evolutionary multiobjective optimization using Non-dominated Sorting Genetic Algorithm II (NSGA-II).
• Integrated AlphaFold2 and ProteinMPNN confidence metrics for objective space definition.
• Employed a mutation operator (ESM-1v and ProteinMPNN) for redesigning unfavorable positions.

Key Points:

• Demonstrated significant reduction in bias and variance for RfaH native sequence recovery.
• Outperformed direct application of ProteinMPNN in a fold-switching protein case study.
• Attributed improvements to an informative mutation operator, iterative design, and Pareto front approximation.

Conclusions:

• The evolutionary multiobjective optimization approach offers a versatile framework for complex protein design tasks.
• This method provides optimal design candidates representing diverse trade-off conditions.
• Anticipated broad relevance and adaptability to various models and specifications in protein design.