DALL-E Inspires Virtually Countless New Drugs That Haven’t Existed Before

AI generators may leave a bitter taste in their homeground of art, but bitter is good in the medical field, which could end up reaping a whole world of undiscovered marvels thanks to the concept of tools like DALL-E.

This form of generative artificial intelligence, called a diffusion model, is making waves in biotech labs to dream up designs for never-before-seen proteins, as reported by MIT Technology Review.

Incidentally, two labs introduced their own versions of diffusion models on the same day last week, says the report. There’s Boston-based Generate Biomedicines, which calls its ‘Chroma’ program the “DALL-E 2 of biology,” and the University of Washington’s ‘RoseTTAFold Diffusion’ led by biologist David Baker. Brian Tippe, one of the creators of RoseTTAFold, says their technology has created proteins with no similarity to those in nature.

Granted, diffusion models aren’t new in the industry, but the two labs are the first whose projects aren’t just proof-of-concept work but have real-world applications that are highly versatile.

Chroma, for instance, has generated proteins in the shapes of the 26 letters of the alphabet, as well as resembling the numbers zero through 10. RoseTTAFold is most proud of its generative protein that binds to the parathyroid hormone, which is responsible for maintaining the calcium levels in blood.

Baker says the model was simply instructed to develop a protein that binds to the hormone, and it did it. When the never-before-seen protein was tested in the lab, it attached to the hormone more tightly than existing drugs, as well as any other drug produced using other digital methods.

The generators can conjure up protein designs with specific shapes, sizes, or functions, enabling scientists to discover effective drugs within minutes—when evolution took “millions of years,” revels Gevorg Grigoryan, CTO of Generate Biomedicines.

Ava Amini, a biophysicist at Microsoft Research in Cambridge, Massachusetts, who has studied diffusion models for biotech purposes, explains that what is most exciting about the automation of protein design is how the output complies with “desired constraints.”

And why proteins? you may ask. As MIT Technology Review points out, they’re the building blocks of life—controlling everything from the way we digest food to how the immune system works. This is why proteins are so important in drug development.

With the virtually infinite possibilities that come with AI generation, Baker defines this time as a “technological revolution.”

[via MIT Technology Review and Interesting Engineering, cover photo 263114241 © Iordache Elena Gabriela | Dreamstime.com]