Repurposing Protein Language Models for Latent Flow-Based Fitness Optimization
This addresses the challenge of sparse high-fitness variants in protein design, offering a more efficient method for researchers in computational biology.
The authors tackled protein fitness optimization by repurposing protein language model embeddings into a latent space and using conditional flow-matching to generate high-fitness variants, achieving state-of-the-art performance on AAV and GFP benchmarks.
Protein fitness optimization is challenged by a vast combinatorial landscape where high-fitness variants are extremely sparse. Many current methods either underperform or require computationally expensive gradient-based sampling. We present CHASE, a framework that repurposes the evolutionary knowledge of pretrained protein language models by compressing their embeddings into a compact latent space. By training a conditional flow-matching model with classifier-free guidance, we enable the direct generation of high-fitness variants without predictor-based guidance during the ODE sampling steps. CHASE achieves state-of-the-art performance on AAV and GFP protein design benchmarks. Finally, we show that bootstrapping with synthetic data can further enhance performance in data-constrained settings.