#249 Brice Challamel: How Moderna is Using AI to Disrupt Modern Healthcare

0:00Drug design is not a problem for Moderna. If anything, we design way more drugs than we can bring to market at the moment. So we have more than 50 ongoing clinical trials. You can check them out. They're on our website. What is complicated for us is to go through the clinical trials, to do the medical training, the regulatory approval process. So it's this entire ecosystem for all the right reasons, because we're saving lives, millions of lives, and we need to be very, very thorough on how we do it. But this is probably where AI is going to have the most impact for us now. The AI solutions that we use both to design the drugs, but also to curate the environment, the regulatory environment, the medical environment, the contractual environment, every component of the vast ecosystem of life science and healthcare and pharma, they are products and they have a product life cycle.

0:52They need to be designed, they need to be tested, they need to be adopted, people need to get proficient at them, they need to be improved, curated, and at some point some cases need to be deprecated and replaced with others. There's a growing expense eating into your company's profits. It's your cloud computing bill. You may have gotten the deal to start, but now the spend is sky-high and increasing every year. What if you could cut your cloud bill in half and improve performance at the same time? Well, If you act by May 31st, Oracle Cloud Infrastructure can help you do just that. OCI is the next generation cloud designed for every workload, where you can run any application, including any AI projects, faster and more securely for less.

1:48In fact, Oracle has a special promotion where you can cut your cloud bill in half when you switch to OCI. The savings are real. On average, OCI costs 50 % less for compute, 70 % less for storage, and 80 % less for networking. Join more Skydance Animation and today's innovative AI tech companies who upgraded to OCI and saved. Offer only for new U.S. customers with a minimum financial commitment. See if you qualify for half off at oracle.com slash ionai. That's oracle.com slash ionai. E-Y-E-O-N-A-I all run together. Oracle.com slash Eye on AI. Hello, Craig. My name is Bryce Chalamet. I am Head of AI Product and Innovation at Moderna, the biotech company.

2:58Prior to Moderna, I worked at Google Cloud for five years as Global Head of Transformation, which really entailed designing with corporate executives the future stage of their evolution with technology. Prior to that, I was the CEO of my own company dedicated to innovation and transformation. I committed four books in French, so you'll never know how bad they are, but they exist. You would lend them on Amazon. And prior to this, again, I was a consultant at the Boston Consulting Group and I worked in marketing in a consumer brand called L 'Oréal. My background is in business school and the core of what I do is to combine strategic business needs with technology understanding to drive change.

3:49And that change can be within a paradigm, which we've mostly lived in the last 10 years within the paradigm of cloud, SaaS software, online collaboration. or it can be through a change of paradigm in which case transformation is the name of the game and no longer innovation right think of transformation it's almost three words in one trans is from one situation to another through a barrier think about transparency through class translation through language transcendence from this world to the next right form is both your organization but also your culture think about the formal dinner so this is from one culture and organization to another through a barrier of change and action is always process and result for instance if you think of evolution it's a process as demonstrated by darwin right darwinian evolution also it's an outcome i can show you uh this pen and say hey look at this it's a fantastic evolution.

4:52It's an outcome. So transformation is the continuous process with continuous outcome of evolving from one stage of our existence to the next through a barrier of change. And that's what happens when a full new paradigm is emerging. And I deeply believe that that's what's happening right now with the access revolution to AI. Yeah. And when you talk about head of AI product, is that what you said? Yes. For Moderna. Moderna is, I mean, AI is in the drug discovery process, but does Moderna have AI products that it's selling externally? So that's a great question. we don't sell at the moment it's only AI products.

5:51It might happen at some point in the future the same way that Amazon ended up selling Amazon Web Services which was not their initial product which was selling books. But that took decades and a very specific strategy orientation. But at the moment we don't have at Moderna. We sell drugs, therapies. piece. However, the AI solutions that we use both to design the drugs, but also to curate the environment, the regulatory environment, the medical environment, the contractual environment, every component of the vast ecosystem of life science and healthcare and pharma, they are products and they have a product life cycle.

6:38They need to be designed, They need to be tested. They need to be adopted. People need to get proficient at them. They need to be improved, curated. And at some point, some cases need to be deprecated and replaced with others. And so we really try to have a great product mindset around AI solutions at Moderna. I'll explain why. if you are as we are in the perspective that AI is a utility, that everyone should have it as they have electricity, as they have laptops, or as they have the internet, then they are going to start generating AI products. You call them GPTs, you call them agents. And let's just focus for two seconds on an obvious one, which is the GPTs, right?

7:28Those encapsulations of instructions, knowledge components, and delivery which you share with other people in the company. Let's say you're in the travel and expense team and you have a travel and expense GPT. If for any reason you forget to upload the new travel and expense policy, but people refer to the GPT that you produced for their next trip, then something unpleasant is going to happen. So you need to be a product owner and remember that you have a GPT out there, leveraging AI to reply to people's questions written in plain French, English, German about their next travel and that you need to keep it updated that maybe if the model evolves you to also evolve the instructions that maybe at some point if there's a new solution you need to move on to a new solution.

8:14You see what I'm saying, right? One of the most misunderstood challenges of vast adoption of AI at the workplace is that it turns every knowledge worker into a digital product owner. And that's why I'm head of AI products and innovation. Yeah. And how, across the enterprise, how many products, AI products are there? I mean, you mentioned I hadn't thought of that, the GPT for expenses, I mean, for people understanding those policies. Well, the expense was just an easy example. We have about 1 ,800 GPs in production at Moderna. We probably have as many AI solutions, which are applications calling APIs from different models to a platform, an ML platform, which is called Compute, and our API orchestrator, which is called MChat.

9:16And so the combination of all those products is probably around 4 ,000 to 5 ,000. And the reason why I don't have an exact number for you is that my number one role is to be a platform owner for those products and to make sure that their data is safe and secure, that they are compliant to our own policies and to some degree for those who necessitate with pharma regulation. And so I don't mean to deep dive into each and every single AI product the same way that if you were, say, in charge of the internet, you wouldn't want to deep dive into every time someone searched Google or browsed the net or used a SaaS application.

10:01You would want to make sure, though, that your internet platform has maybe a VPN, has a good bandwidth, is always available for people. You see what I'm saying? Yeah, yeah. You know, AI and big pharma sort of came into the public consciousness with the development of the COVID vaccines. are those solutions were those products that moderna already had developed absolutely we had at the time moderna is an ai native company there are more ways to design an mRNA that calls a protein one specific protein than there are atoms in the universe the same type of challenge as the go game. And the only way to do this properly and to design single-stranded mRNA, which is the only one that is going to work in the cellular environment, that codes reliably and efficiently a given protein out of those trillions of trillions of solutions, requires AI to be design properly.

11:14So no AI, no mRNA. Yeah. Can I ask about what you say? You don't do a deep dive in every product, but on that, and you mentioned, did you mention AlphaGo? But of course, everyone thinks of AlphaGo. Is that reinforcement learning that you use to train the model to find the correct? Yes, so that's a great question. The models that we're discussing right now, the ones that contributed to the origination of the mRNA industry, really, they are machine learning models. I say this because very often now when we talk about AI, we think about generative AI. and generative AI is a fantastic access revolution to a technology revolution that was machine learning.

12:11And just now, instead of having to code your request in Python or in R, you can just type them in English. So in some ways, even though it has a lot of other merits, I think of generative AI as I think of the graphic interface for computers. Before that, you had to type commands in Unix, and then you could just click on the mouse and drag and drop objects, and you had Visual Basic, and the world of computing transformed itself. That's another moment of transformation because it wasn't a few hundred people in the world. It was millions and millions of people who could use it. So generative AI allows you in plain language, even speaking or typing or even with an image, to trigger under the hood an ML process.

13:02So at the time, back in the 2010s when Moderna was founded, AI at the time was mostly machine learning. And it was this five-star hotel at the end of a dirt road. It was hard to get to, hard to come back from, and very, very difficult to implement efficiently. There was the handful of people and use cases. and certainly Moderna was for me like the invention of the radio after the democratization of electricity right compute power was there for a while and then Moderna appears and I was at the time at Google and I knew that I was seeing the first real use case of all that compute power and all those machine learning models and those reinforcement processes that you mentioned and now the next stage of this is well if it is an access revolution how much access should I give access to everyone and how much power behind that access?

13:59And then how do I organize to make sure that they're proficient and that they really embrace the new way of working that goes with it? And so talking about the drug discovery end of this, are there a family of products, quote-unquote, either models or systems that you've developed and are developing that then the scientists that are designing new drugs have access? Yes. And is there like, I imagine there's a dashboard or something where you see all of these available tools? Well, then again, we have an ML platform, which I mentioned earlier, called Compute. All of our solutions point to Compute, and Compute can be called from almost all of our solutions.

14:59We are a platform company at the core, which means, first and foremost for us, a biological platform company. because the mRNA chemistry, those trends of mRNA encapsulated into nanolipid bubbles, they are a platform unto themselves. They can help cure cancer. They can help with immune and inherited ailments. They can work for virology. there is a huge share of the burden of disease that can be potentially either cured or supported with mRNA technology. And when you learn something from, let's say, development in oncology, you can most often reapply that learning to virology or to rare disease.

15:47And that's the property of a platform, is that you have at every given moment an overview of everything that's running there that's interconnected and you have scale effects in the shared learnings and best practices of every component of it and then on the on the side side by side to that platform which the code of life right mRNA is another platform which is a code of software AI and those two platforms combined are the beating heart of Moderna because all of our AI solutions speak with each other, share common data repositories, our AI engineers collaborate with each other to improve the solutions, and it all serves the biology platform in return.

16:40So one of the most interesting point here is that everyone's focus right now is on drug generation, like novel therapies. I'm not sure this is not a little bit misled, because we are already making inroads in drug generation. The way that amaranic therapies are being conceived, if you think of the prior generations of pharma products, which were finding herbal medicines in a corner of Africa or Southeast Asia, looking at the molecule that would actually have an effect, refining it, producing it, or filtering it, putting it in a pill, or, you know. And now here we are generating therapies from software, assembling them as nanoparticles in a production site and pointing them as nanotechnologies to the heart of the cell to inform the way that the cell generates proteins, either to metabolize components in our biological ecosystem or to alert of threats and trigger the immune system or to self-regulate on cellular production.

18:01What an extraordinary different paradigm of pharmacy, of life science, right? And so drug design is not a problem for Moderna. If anything, we design way more drugs than we can bring to market at the moment. So we have more than 50 ongoing clinical trials. You can check them out. They're on our website. what is complicated for us is to go through the clinical trials to do the medical training the regulatory approval you know process so it's this entire ecosystem for all the right reasons because we're saving lives millions of lives and we need to be very very thorough on how we do it but this is probably where ai is going to have the most impact for us now because drug design we've got this.

18:51And we've had this for a while now. And we share a lot of this with the rest of the industry. Our oncology practice, our cancer treatments is done in partnership with Merck. So we have strong industry partnerships to go together to market. But every clinical trial that we can accelerate, every regulatory filing that we can accelerate, every moment of technical development and industrialization that we can accelerate, we can translate in lives saved. And this is where we see the most impact of AI now, incremental impact of AI. And this is why we want to think of AI as a utility and share it with every component of the company and its ecosystem.

19:41Because those companies, they don't have a department that is just there because it's pleasant. Everyone matters. And it is a chain. It's as strong as the weakest link. Yeah. If any of those links is slowing down, the entire chain is slowing down. We can't afford to have 20 % of our people making progress at the speed of AI and 80 % waiting. And we can't afford either, by the way, to have 80 % of our people accelerating at the speed of AI and 20 % on the sideline. Yeah. It needs to be mass adoption, and that's how the entire ecosystem is going to thrive and accelerate the go-to market of this huge pipeline that we have, and that begs to demonstrate itself for clinical trial, to be approved by regulators and governments, and to go meet the patients who are waiting for them.

20:35Yeah, I have a question about that. I had a fascinating conversation with a startup in Australia a few months ago. TrialKey was their name, but their parent company, I'm not going to be able to remember. TrialKey was the name of the product. And they've trained a model to predict the likelihood of a clinical trial moving to the next phase. Because you invest so much in preparing and putting drugs into clinical trials that if it fails the trial, you've wasted a lot of money, a lot of time. Does Moderna have something like that where you don't go into a clinical trial blind because you have to choose which drugs you're going to put forward that you can gain the likelihood or the percentage of success that you can expect from a clinical trial so you don't waste time on drugs that aren't going to make it?

21:44This is a great question. I don't know that we have that kind of model. Sorry. I don't know that we have that kind of model. I will say this. Our technology is very reliable in the way that a huge majority of clinical trials have had the expected outcome.

22:10it was we we couldn't know as a company before we truly engaged how efficient the technology would be mrna was an unproven technology uh 20 years ago we started doing clinical trials we the humans started doing clinical trials with mrna technology in 2001 even though the design dates from the 50s. And it wasn't proven on the market when COVID broke. There was no mRNA therapy on the market when COVID broke. Today, we have two mRNA vaccines, one for COVID and one for RSV. And we are the only company in the world to have two mRNA products. Pfizer also has a COVID vaccine that is based on mRNA, but their RZ vaccine is a traditional vaccine technology.

23:02And so in this, we are very unique and different because Moderna really is an expansion of mRNA, mRNA Moderna, right? We are an mRNA company. This is what we do, the platform. And we tend now to expect a positive outcome at our clinical trials. We tend not to think that we're playing dice and that maybe it lands, maybe it doesn't because it has almost always landed the way that we expected it to. So the consideration for us and our need to predict, the focus wouldn't be on clinical trials at this point. It would more be on the way the market receives the product, on the way that the supply chain can meet the demand, on the way that the regulatory authorities are going to welcome it and ask for more information.

24:00So those are more the ways that we point. One thing that is really great in your question is that we go back to the fundamental reasons to be using AI to begin with. I know five of them. And I've never met another. And I've been doing this for more than a decade now. You can use AI for perception. To perceive something. think about the tumor in the x-ray, right? Or the spelling mistake on your Word document. How do you identify the signal from the noise, right? That's a huge pillar of usage of AI, which we use for drug design, which we use for regulatory submissions, which we use for contracting.

24:40Like perception is extraordinary from AI. Then there's categorization. Think about the playlist on Spotify, right? that know all your music and put music styles. So this can help us understand client clusters, understand types of industrial processes, even understand our own files to some degree and create folders around the files, if you would, right? That wraps them in a more efficient way. The third is prediction. And you just put your finger on it. AI is fantastic at observing trends and predicting patterns based on input and output of those trends. And that's why I said that currently our prediction needs as a company as a whole, because everyone in every department has their own smaller granular level prediction needs, right?

25:31Someone in legal also might want to anticipate the outcome of a litigation. Someone in HR would like to anticipate arrivals or departures of people in the company. You see what I'm saying? So the prediction really, if you take one level down from the corporate perspective, is useful to almost everyone in the company. And that's why we want to democratize AI. But as a company, it is not so much clinical trial as it is go-to-market that is really our focus right now. Like, how do we bring our pipeline to life? And how do we make it save people's lives in the real world? And then the last two would be recommendation, based on this.

26:04So that would be the next song on your Spotify taste. And Generation, which was there for a while. Remember, we had to complete an email and it would propose the next word of your email. And it just proposes the entire email, right? So generation has made leaps and bounds, but was there since the very beginning of machine learning. It just has taken a whole new dimension with generative AI. So one way that we work with our people, because then again, we want to think in a platform way, in a holistic way. We meet with them in workshops, in large environments. and we ask them to think about their reason to be here, their mission.

26:46Then to break it down into what they need to perceive, what they need to categorize, what they need to anticipate, what they need to recommend, and what they need to generate. And draw lists of those things. Everything in those lists is potentially an AI use case. And then we start creating connections between them, putting them in groups of things that work together, the cats with the cats, the dogs with the dogs. And then we go into AI solution design, most of which they can now orchestrate themselves because then again, we have a plain language interface to machine learning, which is generative AI.

27:26And I think that we have not fully yet completely understood the depth and the impact of that revolution of those five superpowers of AI now granted to every single knowledge worker in a holistic environment like modernist. Yeah. I have two sort of directions I want to go. One, before I forget, I want to talk about personalized medicine, but I'll hold that. One of the bottlenecks for new drugs is the regulatory process. So you're confident of the clinical trials. You get through the clinical trials, but then it has to be reviewed and go through an approval process. Do you work, for example, in the United States with the FDA on helping them implement AI tools to speed up that process?

28:29or are they doing that on their own? Because it seems from their side, they should be able to take your clinical trial report and put it through an AI system that's been trained on all the variabilities and can speed up that process. So that's another great question. We have a fantastic partnership with the FDA. they were an amazing partner to Moderna in the approval process of the very first mRNA product ever to go to market in America which was the COVID vaccine in trying times and it created a very strong bond of knowledge and trust and two-way respect between them and us on how we're going to go together.

29:22A lot of things we were doing at the time were very new because then again it's a new paradigm for life science, to design drugs the way we do versus the way they were designed before. And I think there's a lot of, then again, of understanding, of respect that was created in those years. There was very important years for us and I think for the world, really. It accelerated the arrival of a whole new generation of medicine that probably would have taken a decade or more if it hadn't been for the emergency of the COVID pandemic. So we have great dialogue. The FDA has published a draft guidance on AI usage for various use cases in January.

30:06The industry is working on a comment on that draft guidance, and it is made for this. They publish a draft guidance to give a chance to the industry to react to it. the FDA themselves have regularly expressed their desire to accelerate the approval process thanks to AI and I have a lot of admiration for them to do this they are not a conservative reactive force at all, they are true partners in saving lives with novel medicines they want this to happen and in this we are not accelerationists who are meeting incomprehensive partners this is not at all the situation we are a whole new breed of life science design company meeting people who are fascinated by what can be done and want to approach it in the best way to be respectful of the mandate that the people have given them to make them safe which of course is also our top priority but also we are to market as quick as possible.

31:17And I love the dialogue we have with them. They are very open to the ways that AI is going to help accelerate all that regulatory process that you're mentioning. It's also great accelerant for the way that we reply to their requests. They make a lot of requests, as they should. Mostly responding to one of their requests relies on data aggregation. Most of the data we have already, and they're just asking to see it. and AI is a great way, then again, to perceive and to categorize. So we can perceive the data that is needed to answer the request, and we can categorize it to make it easier to ingest by the FDA when we give it back to them.

31:57It would be crazy for us not to use it, both as a company in terms of efficiency, but also in terms of the outcome, because the FDA wants all the right data in all the right categories and order given to them in the most consumable possible way. So we have a lot of AI supporting our regulatory teams in the way that they prepare and that they organize the data and then they reply to regulatory requests, right? So then again, both in the approval of a medicine, but also in the constant monitoring of how the medicine is produced, how our systems are operating, how we're improving them, there is a constant, very intense dialogue between regulatory authorities and Moderna.

32:46And yes, we absolutely use AI to accelerate it. And that's also the FDA's intent, outspoken intent to have it this way. We do research at Moderna. We don't do policy. But I can only observe with you that every administration in this country has made the acceleration of go-to-market for great therapies a priority, every single one of them. There's been a lot of talk because of AI, because of things like AlphaFold, of personalized medicine, precision medicine. Absolutely. What's happening at Moderna in that regard, and what has to happen for that to become a reality where you go see your doctor and they do a, I don't know, I don't know how it works.

33:43They do a DNA profile and they identify a therapy. So in our case, this is very real because we are right now in phase three clinical trials and personalized cancer medicines. So here is HydeWorks. we do a biopsy of the cancer cells and we look at the proteins in those cells that are unusual that are the outcome of the cancer itself these are called neo-antigents we identify them make sure that they are specific enough that if we point the immune system to them it is not going to touch anything else and then we generate an mRNA that codes those proteins and that points immune system to cancer cells.

34:31Why this matters is because if, for instance, you had a cancer that you went through surgery, that you went through chemotherapy, and you've already been through a lot, and my mom is a cancer survivor, so this is very, very concrete and real to me, what we're talking about now. What you need is remission. What you need is for the cancer not to come back. And the way it comes back is most often through dissemination in your body. most often metastasis it's hard to anticipate it's hard to treat through surgery what you really want is to use the power of the immune system to be on the lookout for any cancerous cell that has escaped the surgery and the chemotherapy and that is potentially an agent that is going to trigger the next tumor so this is as individualized as it could be because every cancer is different and even if you had two or three cancers and really i don't wish but there would be two or three completely different cancers so the drug the treatment that we propose there the therapy is not only individual to you as a person it's individual to your cancer as an event