ProtFlow introduces a novel approach to protein sequence design that significantly improves both efficiency and accuracy by leveraging flow matching on compressed protein language model embeddings.

• Reduces the modeling space from 20^L to d-dimensional space (where d<<L)
• Achieves 15-35x faster inference speed compared to diffusion models
• Demonstrates superior performance in designing functional proteins including antibodies and antimicrobial peptides
• Leverages the semantic knowledge captured by protein language models for better designs

This advancement has profound implications for biological research and drug discovery, potentially accelerating the development of therapeutic proteins and antibodies for various diseases.

ProtFlow: Fast Protein Sequence Design via Flow Matching on Compressed Protein Language Model Embeddings