ProSpero: Active Learning for Robust Protein Design Beyond Wild-Type Neighborhoods

2025-05-28

AI Summary

In protein engineering, jointly optimizing sequence fitness and novelty—while escaping the wild-type neighborhood and preserving biological plausibility—remains a major challenge for data-efficient design. This paper introduces an active learning framework that couples a frozen pretrained generative model (e.g., ProGen) with a dynamically updated surrogate model. Our method innovatively integrates fitness-driven residue importance scoring with biologically constrained sequential Monte Carlo sampling, substantially improving robustness under model misspecification. Across multiple benchmark tasks, our approach consistently matches or surpasses state-of-the-art methods: generated sequences achieve high experimental fitness (>95% functional among top-100 candidates) and high novelty (average sequence identity <30% relative to wild-type). The framework enables interpretable, scalable protein design in low-data regimes, establishing a new paradigm for biologically grounded generative optimization.

Technology Category

• Search and Optimization: Sampling/Simulation-based Search
• Machine Learning: Active Learning
• Constraint Satisfaction and Optimization: Constraint Learning and Acquisition

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications

Abstract

Designing protein sequences of both high fitness and novelty is a challenging task in data-efficient protein engineering. Exploration beyond wild-type neighborhoods often leads to biologically implausible sequences or relies on surrogate models that lose fidelity in novel regions. Here, we propose ProSpero, an active learning framework in which a frozen pre-trained generative model is guided by a surrogate updated from oracle feedback. By integrating fitness-relevant residue selection with biologically-constrained Sequential Monte Carlo sampling, our approach enables exploration beyond wild-type neighborhoods while preserving biological plausibility. We show that our framework remains effective even when the surrogate is misspecified. ProSpero consistently outperforms or matches existing methods across diverse protein engineering tasks, retrieving sequences of both high fitness and novelty.

Problem

• Designing high-fitness novel protein sequences efficiently
• Exploring beyond wild-type without biological implausibility
• Improving surrogate model fidelity in novel regions

Innovation

• Active learning with frozen generative model
• Fitness-relevant residue selection integration
• Biologically-constrained Sequential Monte Carlo sampling