UNITED KINGDOM - DECEMBER 23: Raindrops on glass (Photo by Tim Graham/Getty Images)
It seems ridiculous to think that you could reduce something like amino acid interactions and resulting proteins to the level of something like Minecraft.
In fact, I wouldn’t think most of us have even considered how generative AI could be applied to life science.
But a few people behind the scenes are working diligently on figuring this out.
Geoff von Maltzahn is at Flagship Pioneering, a company that’s working on the vanguard of high-tech life science. This venture capital firm has teams of scientists and entrepreneurs inventing new technologies like mRNA and novel antibody therapeutics, as well as broader genomic projects. Flagship Pioneering created companies like Moderna Therapeutics, Seres Therapeutics and Indigo Agriculture, to innovate in places where breakthrough scientific progress can help with the biggest challenges that we as humans face.
In a compelling TED talk, von Maltzahn illustrates how all of this can work.
range
He starts with a basic explanation of protein folding, where individual proteins emerge through a process of evolution that biologists tend to understand fairly well, while the rest of us have don’t have much of a clue.
The World of Proteins
Part of what I thought was so great about this talk was that von Maltzahn really breaks down formidable scientific ideas into words that many people can easily understand.
For example, in showing how a protein structure evolves, he notes that each protein starts out as a ‘noodle’ and then folds itself into useful forms.
“This is the equivalent of, if your car, your refrigerator, your house, arrived as a noodle and folded itself into the form that you enjoy it in,” he says.
After explaining protein folding, he explains that new systems are able to generate models of proteins that fold in very precise ways. Von Maltzahn talks the prospect of about pursuing a ‘Shakespearean mastery’ of protein writing.
His metaphor on precision and archery, again, boils down this type of innovation into a sort of visual reference in explaining how generative AI models can create far improved antibodies for existing therapeutic targets in inflammatory diseases, cancer, and more:
“It’s kind of like Robin Hood splitting the first arrow by firing an arrow right down the center of it, 50 times in a row,” he says. “And the reason that it has the potential to be valuable is: every time an antibody shows … that it can provide a benefit to patients, humanity now has the ability to access the very best antibody [with this technology].”
In aid of talking about the best antibodies, he also goes over specifically how their team applied generative AI to discover antibodies that can neutralize the Covid virus that created the universal pandemic we all suffered through. He explains that to keep viruses from entering the cells, the technology can build antibodies that go after that portion that’s attaching to receptors and involved in cell entry. Because these domains haven’t evolved very much between the original Covid viral strains and today’s wildest variants, they offer the potential longer-lived protection from infection.
“This may be important to the next pandemic,” von Maltzahn says.
More New Science
In addition, von Maltzahn talks about creating new enzymes and analyzing things like the ‘tree of life of SARS viruses’ and how that will lead to very profound discoveries.
“Maybe generative AI isn't just going to create beautiful cartoons and limericks to entertain ourselves and one another, and access knowledge that is already resident within humanity,” he posits. “It may just expand our access to new knowledge and new technologies in important ways.”
How big will this be?
The space of unexplored biology he explains is vast. Mentioning an analogy to the ‘great age of exploration’ with ships, he says we have worlds of knowledge to discover.
“Even with generous assumptions around everything Mother Nature has ever had the chance to build and test throughout the entire history of evolution, all of her lab has tested less than the expanse of one drop of water relative to all of the Earth's oceans of possible protein sequences,” he says. “Think about that for just a second. Everybody in this room, every protein that makes our lives possible, every one of our ancestors and everything else in the living world that has ever been, fits into that one drop of water. … if everything we know is in that one drop of water, if we find one more drop of water, that'll be a really big deal. But there really isn't a reason to believe that it's going to be limited to that. Just imagine what could be.”
I’ve seen and heard a lot about AI in the past year, but this is something fairly unique that really deserves a close look. We’re just beginning to understand how AI can create digital worlds that are vibrant, unique, and compelling, and how it can think like humans. But being able to create in a godlike way, using the building blocks of all life– well, to put it simply, that that’s something else.
