Sometimes when you’re considering how to bring the power of AI to a clinical context, it sort of takes a new way of thinking to get inspired about what’s possible.
I was thinking about this the other day, inspired by some people who have been working hard on genomics, oncology research, and other types of biological and anatomical applications. There’s so much of it, suddenly, especially at these institutions that I’m so close to – to call it a “revolution” in my view, isn’t hyperbolic.
So what about a new way of looking at medical AI? To me, Ava Amini at Microsoft Research really articulates this kind of framework and this kind of goal when it comes to, in her words, “precision medicine” and “precision health,” and the marriage of biology and AI for good. She has described the effort this way:
“Our work is really fundamentally grounded in the fact and the realization that human life is a marvel: on the one hand, incredibly robust, and on the other, incredibly delicate. … We don’t just need a better way to detect disease, or discover a new drug – I’d argue that we need a whole new way of thinking, a way to leverage data about the human body, about biology, and learn these patterns naturally and intrinsically.”
Our bodies, she suggests in panel talks and lectures, are unique organisms that work together, but a lot of medical applications don’t take our uniqueness into account.
Too often, we are treated, as she points out, like an “instance in a population” instead of a unique human being. And of course, clinical science is data-driven, and that’s a good thing, but assuming that all patients are (X) is sort of oversimplifying things and maybe making us miss something. In fact, I know a lot of good doctors who would clam this assertion, too, saying that you start with the medical journals, but you also meet your patient where they are.
So what happens when you personalize medicine?
Well, you get better targeted outcomes for individual people. Patients feel better served – doctors feel more competent and successful.
I’ll lay out some numbers that I heard Amini mention as part of her presentation recently, that with one biopsy you get:
· 100,000 mutations
· 10 million genes
· 200,000 proteins
· 34 million cells
That sort of puts it in perspective, doesn’t it?
I was also excited about this project that’s happening with not only Microsoft, but also both MIT and Harvard as stakeholders, that will help to enhance the pipeline and get more data into clinical AI work.
Amini describes it as a “flywheel” to generate new data that is tested in the lab and refined for clinical practice. I like that metaphor!
There is this tool EvoDiff that I heard about from Amini, too: what is it? Well, GitHub explains that it is “a framework that combines evolutionary data with diffusion models for controllable protein generation in sequence space.” It’s also now open source, which is very cool, in this environment where communities are trying to work together to leverage AI, rather than keep tools in walled gardens.
To me, that’s a very effective way to frame this kind of work – unlocking the power of biology with genAI and artificial intelligence tools. Yes, we use AI in radiology. Yes, it plays a role in diagnosis. But it’s also much more of a Socratic helper when we use it to form better ideas about each of our unique bodies and how to treat them!
