<ul data-eligibleForWebStory="true"><li>ProtBFN is a 650 million-parameter foundation model for protein sequence design introduced in a recent paper in Nature Communications.</li><li>It utilizes Bayesian Flow Networks to generate diverse protein sequences without explicit structural data, offering unconditional and conditional protein generation.</li><li>The model outperforms leading autoregressive and diffusion models, produces sequences matching natural length and amino acid distributions, and includes a fine-tuned variant for antibody heavy chains.</li><li>ProtBFN's approach enables zero-shot design, making it versatile for therapeutic and industrial enzyme design, with open-source availability for benchmarking and community contributions.</li></ul>

# ProtBFN: Bayesian Foundation Model for Protein Sequence Design

Discover more