Microsoft has introduced a new framework called EvoDiff that can generate diverse proteins given a protein sequence. Unlike the current methods for protein design, EvoDiff does not require any structural information about the target protein. It eliminates the most laborious step of the process, making protein engineering more efficient and cost-effective.
EvoDiff is a diffusion model trained on data from different species and functional classes of proteins. It gradually subtracts noise from a starting protein, transforming it into a protein sequence. The framework can be used to create enzymes for therapeutics, drug delivery methods, and industrial chemical reactions. It can also fill in gaps in existing protein designs and synthesize disordered proteins that play important roles in biology and disease.
The research behind EvoDiff has not yet been peer reviewed. However, the Microsoft team plans to test the proteins generated by the model in the lab to determine their viability. They also aim to scale up the framework and achieve more fine-grained control by conditioning EvoDiff on text, chemical information, or other factors that specify the desired function.
Overall, EvoDiff has the potential to revolutionize protein engineering by shifting the focus from structure to sequence. By utilizing this framework, researchers can design new proteins with greater efficiency and expand the capabilities in protein engineering.
– TechCrunch: [link]
– Microsoft Research Blog: [link]