Steering Protein Family Design through Profile Bayesian Flow

• DOI: 10.48550/arxiv.2502.07671
• Publication Date: 2025-02-11
• Authors: Jingjing Gong; Pei Yu; Siyu Long; Yuxuan Song; Zhe Zhang; Wenhao Huang; Ziyao Cao; S. H. Zhang; Hao Zhou; Wei‐Ying Ma
• Keywords: Quantitative Biology - Biomolecules; FOS: Biological sciences; Biomolecules (q-bio.BM)

Protein family design emerges as a promising alternative by combining the advantages of de novo protein and mutation-based directed evolution.In this paper, we propose ProfileBFN, Profile Bayesian Flow Networks, for specifically generative modeling families. ProfileBFN extends discrete Network from an MSA profile perspective, which can be trained on single sequences regarding it degenerate profile, thereby achieving efficient avoiding large-scale data construction training. Empirical results show that has profound understanding proteins. When generating diverse novel proteins, accurately capture structural characteristics family. The enzyme produced method is more likely than previous approach to have corresponding function, offering better odds proteins with desired functionality.