SAGE-prot: scoring-assisted generative exploration for multi-objective protein design.

Area of Science:

• Biotechnology
• Protein Engineering
• Computational Biology

Background:

• Designing proteins with multiple optimized properties is challenging, especially with conflicting objectives or lack of structural templates.
• Existing methods often rely on structural information or lack efficient optimization strategies for complex design tasks.

Purpose of the Study:

• To present SAGE-Prot, a framework for optimizing protein sequences directly, integrating generative models and genetic algorithms.
• To demonstrate SAGE-Prot's effectiveness in improving protein properties like binding affinity, stability, enzymatic activity, and solubility.

Main Methods:

• SAGE-Prot employs autoregressive sequence generation, genetic algorithm (GA)-based diversification, and scoring-guided property evaluation in a closed-loop system.
• It optimizes protein sequences directly without templates, enabling structure-aware evaluation.
• A curriculum learning (CL) strategy was used to accelerate convergence and enhance design quality.

Main Results:

• SAGE-Prot, using hybrid language model/GA strategies, outperformed baseline methods in rediscovery and similarity benchmarks for 10 therapeutic proteins.
• The framework successfully identified high-performing variants for protein G domain B1 and TEM-1 β-lactamase.
• Experimental validation showed up to a 752-fold increase in catalytic activity for designed TEM-1 β-lactamase variants.

Conclusions:

• SAGE-Prot provides a generalizable, data-driven approach for protein engineering across diverse optimization landscapes.
• Coupling deep generative modeling with structure-informed evaluation and iterative fine-tuning enables efficient discovery of functional protein variants.
• This framework has significant practical utility for advancing biotechnology and therapeutic protein design.