Microsoft Open Sources AI Protein Design Model to Enable Faster, Cheaper Protein Engineering

Microsoft open sourced EvoDiff, an AI model that can generate novel protein sequences without needing structural information. This makes protein design faster and cheaper.
EvoDiff is a diffusion model trained on protein data to gradually create protein sequences from noise. It allows controllable protein design focused just on sequence.
The model can not only create proteins, but also fill in gaps around parts of known proteins. It can make disordered proteins too.
EvoDiff is unreviewed for now. More scaling work is needed before commercial use. The team plans to test generated proteins in the lab.
If lab tests confirm viability, the next version will focus on more fine-grained control over function using text, chemical data, etc.

Microsoft open sourced EvoDiff, an AI model that can generate novel protein sequences without needing structural information. This makes protein design faster and cheaper.
EvoDiff is a diffusion model trained on protein data to gradually create protein sequences from noise. It allows controllable protein design focused just on sequence.
The model can not only create proteins, but also fill in gaps around parts of known proteins. It can make disordered proteins too.
EvoDiff is unreviewed for now. More scaling work is needed before commercial use. The team plans to test generated proteins in the lab.
If lab tests confirm viability, the next version will focus on more fine-grained control over function using text, chemical data, etc.