It’s the Wild West with the advent of Diffusion models, say protein designers. Diffusion models are what allow AI to create images using text prompts, such as with DALL•E, but as it turns out, similar models are now designing novel proteins. Thus, AI can build an impressive and creative painting or design something that could be in your body.
Biotech labs are already using the AI to invent proteins in the pursuit of new drugs, saying they can do in minutes what evolution has taken millions of years.
“If you’re doing something with deep learning and you’re not working on diffusion models, you’re left behind at this point,” Nicholas Polizzi, a protein designer at Dana Farber Cancer Institute, tells the Boston Globe. “It’s sort of a Wild West with diffusion models right now, because I think everybody knows this is the next big thing in protein design.”
To some protein designers, proteins are like the “machines of biology,” a little strange to think about because that analogy means much of our bodies are made up of those machines. That’s how Michael Nally of a Boston-based startup called Generate Biomedicines described it. MIT Technology Review says they liken their “Chroma” program to “the DALL•E2 of biology.”
“Proteins are the machines of biology — the motors, the sensors — that make everything go,” said Nally. “We can now create these machines in whatever shape we need to drive the desired biological task.”
Protein Machines Designed by AI
As the fundamental building blocks of living things, it might seem odd to think of proteins in a mechanical sense. After all, our genes encode the amino acids that build chains of proteins. And it does make you wonder why this type of AI would stop at just making proteins. If it can do that, could it take it all the way to building entire organisms, for example? Well, for now, one of the primary goals is stopping viruses, creating proteins that efficiently bind to them (And, of course, big profits from the resulting new drugs).
New proteins could help solve many problems, such as infant mortality, fighting cancer, creating universal vaccines, reducing chronic pain, and improving digestion problems. They could even create next-generation materials and help solve ecological issues (see the video below).
More about designing proteins with David Baker of the Institute for Protein Design below via TED (from 2019):
Generating Proteins Using AI
Looking at the mesmerizing Chroma website, one is immediately reminded of DALL•E 2, except now we’re generating molecules, not pictures. Thankfully, you can’t sign up to start doing it yourself, but more on that later. They say Chroma can “directly generate large molecular assemblies at the scales frequently seen in nature.” If so, then we’ve reached the point when AI is designing what will be in nature, like it or not.
On the other hand, the program also created molecules shaped like the Latin alphabet and numbers.
Meanwhile, other researchers are using similar programs to design and bring new proteins to life, such as University of Washington researchers using a program called RoseTTAFold Diffusion.
The Institute for Protein Design in Seattle shared that others are designing protein “Lego Blocks,” enzymes, and “self-assembling semiconductors.”
Designing Proteins with Video Games?
Several years ago, video gamers were already designing proteins. For example, researchers from the University of Washington used a computer game called Foldit to allow “citizen scientists” to design new synthetic proteins (see below).
They found that it didn’t take a biochemist to design proteins, but people who enjoyed an interesting puzzle – kind of mind-boggling, humbling, and a little alarming to think about.
Researcher Brian Koepnick said the team hoped to find new vaccines or cancer treatments, saying, “We don’t want to wait billions of years for nature to evolve proteins. We’ve shown that Foldit players, these non-experts, can design proteins successfully and just as well as anyone at the Institute would.”
Who knew that playing a video game could lead to real-world science that could have value, or in some cases, maybe create new problems? Now, what happens when Diffusion models are integrated into the game? Do we want to know?
Video by UW Medicine below:
Featured image created with DALL•E 2