Property guidance for protein sequence generative models with ProteinGuide
Nature Biotechnology, Published online: 29 July 2026; doi:10.1038/s41587-026-03207-z On-the-fly conditioning of pretrained protein generative models guides protein generation toward specific properties.
ProteinGuide is a novel method that allows for conditioning sequence generative models on auxiliary information, such as experimental data, without the need for additional training. This framework is applicable to a wide range of protein generative models, including masked language models, autoregressive models, diffusion and flow-matching models, and more.
The authors demonstrate the effectiveness of ProteinGuide by using pretrained generative models to design proteins with specific properties, such as increased stability or activity. Additionally, the method can optimize for two desired properties that may conflict with each other. The researchers also showcase the application of ProteinGuide in conjunction with wet-lab data generation, resulting in improved editing activity of a base editor in vivo.
The code, data, and tools associated with ProteinGuide are made available to the public, allowing for easy access and further exploration of this innovative approach to protein design.
Written by urgent.news from Nature Biotechnology's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
