A Variational Perspective on Generative Protein Fitness Optimization

2025-01-31

AI Summary

Protein fitness optimization constitutes a high-dimensional discrete sequence search challenge. Method: This paper introduces the first variational inference–based protein design framework, which maps amino acid sequences into a continuous latent space and jointly models a mutation prior—implemented via flow matching—and a fitness likelihood provided by a predictor. Optimization is accelerated through PLM-derived embeddings and latent-space gradient guidance. Contribution/Results: The framework explicitly decouples prior and likelihood, enabling plug-and-play customization for diverse design objectives. Evaluated on two protein benchmarks with markedly different complexity levels, our method achieves state-of-the-art performance, significantly improving both the sampling efficiency and success rate of high-fitness variants.

Technology Category

• Search and Optimization: Sampling/Simulation-based Search
• Machine Learning: Optimization
• Reasoning under Uncertainty: Relational Probabilistic Models

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

The goal of protein fitness optimization is to discover new protein variants with enhanced fitness for a given use. The vast search space and the sparsely populated fitness landscape, along with the discrete nature of protein sequences, pose significant challenges when trying to determine the gradient towards configurations with higher fitness. We introduce Variational Latent Generative Protein Optimization (VLGPO), a variational perspective on fitness optimization. Our method embeds protein sequences in a continuous latent space to enable efficient sampling from the fitness distribution and combines a (learned) flow matching prior over sequence mutations with a fitness predictor to guide optimization towards sequences with high fitness. VLGPO achieves state-of-the-art results on two different protein benchmarks of varying complexity. Moreover, the variational design with explicit prior and likelihood functions offers a flexible plug-and-play framework that can be easily customized to suit various protein design tasks.

Problem

• Protein Design
• Optimization
• High-Dimensional Search

Innovation

• VLGPO Method
• Protein Adaptability Assessment
• Flexible Protein Engineering