Cerebras SUPERNOVA 2025 | Dr Shant Ayanian, Mayo Clinic

Clips
More from Cerebras SUPERNOVA 2025

Dr. Shant Ayanian

Staff Consultant

Mayo Clinic

play_circle_outline Enhancing AI Efficiency Through Genomic Models for Personalized Medication Response in Autoimmune Diseases

Info
Transcript

Dr Shant Ayanian, Mayo Clinic

Dr. Shant Ayanian

Staff Consultant Mayo Clinic

Shant Ayanian, M.D., staff consultant at the Mayo Clinic, joins theCUBE’s John Furrier at Cerebras SUPERNOVA 2025 to explore how AI and genomics are converging to reshape clinical medicine. Speaking from Fort Mason in San Francisco, Dr. Ayanian highlights how Mayo Clinic’s collaboration with Cerebras is accelerating the development of genomic models that improve diagnostic precision and personalize treatment pathways.

The conversation explores the application of digital twin methodologies to predict personalized medication responses, particularly for... Read more

explore Keep Exploring

What is the involvement with Cerebras and genomic models in clinical practice? add

bolt Powered by CUBE AI

Dr Shant Ayanian, Mayo Clinic

search

>> Hello, I'm John Furrier with the theCUBE. We are here in San Francisco, Fort Mason, for the Cerebras Supernova event. It's turning into a conference, oversubscribed. A lot of people are going to be rolling in over the next few hours, and a big party tonight. Of course we've got all the action. Dr. Shant Ayanian, staff consultant, Mayo Clinic. Doctor, good to see you.

>> Thank you.

>> Thanks for coming on.

>> Thank you.

>> So Cerebras, they have the fastest reasoning, fastest tokens out, fastest tokens. Obviously the AI infrastructure world loves that. It enables more speed, more tokens means more analysis, more goodness from the AI. So tell us your story and your involvement with Cerebras.

>> Yeah, I think the goodness for us comes from our genomic models. And so what we're trying to do is really take all of our patients' genomic information and create a brand new model that's able to be integrated into the clinical practice. We need the model to be fast. We need the model to be effective but also efficient. So we found that Cerebras would be the perfect company for that. We've been working with them for the past 18 months at this point. We have two models already up and running. We're testing them on our actual patients right now in a study, of course, in a research study, trying to understand if we can predict specific response to medications. So if I have a patient who comes in with a specific autoimmune disease, what is the best medication I can treat them with that would be able to treat them fast and effectively?

>> This is a great area, because there's been a lot of buzz around digital twins.

>> Correct.

>> And simulations. This is where it's actually working. This is where you guys are pushing the envelope on this. Is that right? Am I getting that right?

>> That's absolutely right. This is the next iteration of what a digital twin should be. We're not looking just at their digital fingerprint, meaning their EHR, their x-rays, or their imaging. We're actually looking at their genomes, their actual genes, and trying to understand how we can recreate some of that in silico, within the computer, basically.

>> This is a breakthrough. I just want to riff on this a little bit because I want to understand the scope of the magnitude. Because in the old days I would talk to physicians and experts and be like, "Yeah, we have a model that can just general purpose, but some corner cases are hard to do." One of the promises of AI is personalization.

>> Correct.

>> Is this where you're going with this? Obviously every person's different, right? Can I have my own?

>> Yes.

>> And you can model me up?

>> And that's exactly what we're doing. Every patient will have their own genome and every patient will have their own embeddings, and we would integrate those embeddings, the genomic embeddings, along with all of their other embeddings. And that would create that multimodal single individualized model that can predict whatever it is that you want to predict.

>> It's like a health check predictor. "Let's test the antibiotics on John." Or when everyone gets a procedure, they always have to get anesthesia or all kind of testing, or five zillion questionnaires. What did you miss? I don't know what I'm allergic to. I get older, maybe I'm allergic to something. So this is like the future. How do you guys set this up? Take us through the layout. You just roll in some Cerebras and give some blood? Obviously a little bit over the top, but take me through the mechanics of how you guys do this.

>> We're lucky at Mayo Clinic, because we have a lot of this data already ready prepared. We have one of, I think, the best IT infrastructure in the world. So we've had the digital capabilities of creating something like this. We just needed a compute partner, and that's where Cerebras would come in. We've used data that's been collected over the last 20, 30 years, even actually going back to the last 90 years, and we've been using that data to really create these models.

>> It's almost like we used to have an expression 10 years ago when big data was kicking on the scene, or 15 years ago, data hoarders and data scroungers. Now people, if you get your hands on the data, you can actually do the work. Take us through what that enables for you guys to do. Take me through like, okay, is it up and running right now? What's the status of the project? You said you're doing some testing with users?

>> It's actually, yes, and it's in the process of becoming basically a clinical trial at this point. And what we mean by that is we want to evaluate the performance of the model in the real world prior to generalizing its use. And the model is not for us. The model is for our patients. At the end of the day, this model needs to reach the patients who need the care the most.

>> What's the involvement of the patient? What's their level of engagement that you need from them?

>> Really nothing.

>> Nothing.

>> Whenever they come to see someone who has an autoimmune disease, let's say, all they need to do is show up and then we will take care of the rest.

>> Got it, so you guys go to the back end. All right, so I have to ask more of an internal, and when you have these changes, it's always people, process, technology, that kind of cliche. Here, I've got to ask, are people excited? Because I can almost imagine the nerd factor of getting supercomputing capability at the fingertips versus signing big fat checks for science projects. The old way was, get an HPC system, roll the dice.

>> Hope it works.

>> Hope it works, or the results are old. You're talking about real fast, real time results.

>> Yes. We're talking about results that will be available within 48 hours for the patient and for the clinician to review. Everyone is excited about this. We just gave another talk two days ago and people were just, "When can I use this? When can I use this for my patients?" We have to take it slow. We have to really evaluate and validate every step of the way, but we're pretty confident.

>> I have to ask you about structural questions. I'm just curious. Maybe it's a dumb question, but we'll see. The old world was always processes, clinical trials, approvals, compliance, documentation reviews, because the consequences were serious, whether it's an implant for the heart or whatever. What does it do to you guys when you start getting this kind of fidelity in the data? Does it change process? Do you hire Gen Zs to come in because they prompt better than the old Boomers and the Gen Xers? You're smiling.

>> It actually is the same exact process. We still have to go through the same exact rigor that any other study would have to go through, any surgical procedure would have to go through. We have to complete all the paperwork. So the paperwork is not any less than if you were to design a brand new medications or get them into the clinical practice. And also, it's not just the prompting, it's the entire infrastructure that needs to be supported. So yes, we did hire a lot more just to get that engineering factor to support this. But yes, the paperwork is still the same.

>> Are people excited?

>> Yes, they're very excited.

>> They're jazzed probably, I can imagine. Almost like candy, kid in a candy store. How about Cerebras? Take me through the relationship there. They have great speed. They're putting out more inference, first token out matters. Take us through some of the implications of that capability that they have and what their advancements mean for you.

>> I think the faster they get, the faster I get my answers, and that's very important. For us, every hour, every minute, every second will count in a clinical practice. So the faster they get, the better for our patients.

>> Yeah. Do you envision a day where this gets an interface to an app? What's the vision? Take us through some of the mindsets of, okay, as you guys ... Because with AI, one of the things I've noticed is that the observation is, once someone has these breakthroughs, new things come out of it. Can you share a story about how you scratch a surface, you poke at it, and then all of a sudden new collaboration happens, people come around problems faster? Take me through an example you can share.

>> So we started with just one disease, which was rheumatoid arthritis, and now we have at least seven diseases within the last six months, seven other conditions that we are starting to work on. People get curious about the power that models like this will offer to their patients, to their clinical practice. So we went from one disease, one treatment, into multiple diseases, multiple treatments, and that's the speed with which you can go.

>> The innovation is contagious.

>> Yes. Yes.

>> Unintended. What's the coolest thing that you've seen, if you had to zoom out? It could be tech, it could be outcome, could be a patient result. What's the coolest thing you've seen?

>> I think for me it's been the people. The people that I work with, and it's those interactions that really make me understand what we're doing, make me understand the power that AI can offer, but also make them understand that what they're doing is going to touch people's lives. They just get so excited about it.

>> Yeah, I think it's a phenomenal breakthrough. I have to ask you the tech question around, obviously there's a lot of conversations around the tech industry, tech for good. We see a lot of tech for good. This is another example you guys are working on as amazing work. And I believe that we're in an era of, the innovation is contagious and spawns more innovation. What's the tech for good angle? If you had to say, "Is this good for people?" Obviously it is. This is actually one of those stories where it works.

>> Yes. And the hope is that this generalizes to cancer, this generalizes to strokes and heart disease and diabetes and chronic diseases. This will generalize to almost every disease that we know.

>> And what are some of the blockers that holding you back? Is it just staff? Is it just more horsepower? Or is there?

>> I think all of that is unfortunately blockers that we would have to go through. It's also just the amount, the number of patients, the number of cases that we would have. These are the little blockers that we face.

>> And you're happy with Cerebras?

>> Oh, yeah. We're happy.

>> The results are good. Are they the fastest?

>> So far, so far.

>> Speed wins. Doctor, thank you for coming on theCUBE.

>> Absolutely.

>> I appreciate you stopping in to our little pop-up cube. Appreciate you coming in. Thanks for coming in.

>> Absolutely. Absolutely. Thank you.

>> I'm John Furrier here at theCUBE. We're at Supernova. This is Cerebras' big event about news, but it's actually turning into a mini conference, over 2,000 people coming in. Don't forget to subscribe, all the coverage here. We'll do whatever it takes to get to that data and share it with you. I'm John Furrier, your host of theCUBE. Thanks for watching.