ProGen

ProGen is a 1.2 billion-parameter protein language model that generates functional, de novo protein sequences across diverse families.

Developed by Salesforce Research and Profluent, ProGen applies large-scale generative AI to the language of evolution. By training on 280 million protein sequences, the model learns the structural grammar of life, enabling the design of synthetic enzymes like lysozymes that match the catalytic efficiency of natural variants. It uses conditional tags (taxonomic family or molecular function) to steer generation, allowing engineers to bypass slow directed evolution cycles and move straight to high-activity biological candidates.

https://progen.ai