Big Pharma, Moderna on An AI-Empowered Era of Drug Discovery

November 8, 2023

By Allison Proffitt

November 8, 2023 | At a panel early last month at the HLTH 2023 event, speakers from Takeda, Bristol Myers Squibb, Moderna and Eli Lilly discussed what moderator Jessica Federer, former Bayer Chief Digital Officer, called a drug discovery renaissance.  

Fueled by technological advances, many in artificial intelligence and machine learning, drug discovery and development look far different than it did even 15 to 20 years ago. If you were starting drug discovery today, Federer asked the panel, what would it look like? The question represented a hypothetical fresh start for the three companies on the panel who count their company histories with three-digit numbers. 

Drug discovery today is a computational problem, said Greg Meyers, Executive Vice President, Chief Digital & Technology Officer, Bristol Myers Squibb, and he challenged the audience to think about processes from a data-first perspective. Data is not simply exhaust from your experimental machine, he said. The data are valuable and being able to effectively mine data generated in the past would be infinitely more valuable than working forward with a blank slate. All drug discovery companies will be technology companies in the future, he predicted. Building a drug discovery process that views data as fuel will be essential to success.   

Ann Heatherington, R&D Chief Data & Technology Officer and Head, Data Sciences Institute Takeda Digital Ventures, focused her vision on drug development and shared Takeda’s current operating mantras: automation first, quality by design, and Takeda processes, Takeda people. These maxims have moved Takeda forward by baby steps so far, she said, but the tipping point is still to come. One of the practical impacts of these ideas is the decentralization of classical IT.  “All of the technology that supports R&D, in my case, lives in R&D,” she said. “What that enables us to do is… to move fast in the way that we're approaching things.” 

Diogo Rau, Executive Vice President and Chief Information and Digital Officer, Eli Lilly, focused on patients first. “For a century or more the way that we've been doing medicine is as if the patient is really just kind of an extra there,” he observed. He rejected the fixed dose, fixed frequency model of drug delivery and argued for a more personalized approach. “What I would really love to do is focus everything on the patient and then work backwards. I think it’s kind of crazy that in this day and age we still follow the same model of prescriptions that we did 50 or 100 years ago. It’s, ‘Take two pills every day for the rest of your life,’ or whatever it is, and it doesn’t really vary based on age, gender or anything else,” he said.  

Finally, Brad Miller, Chief Information Officer, Moderna—representing the only company on the stage with less than a century of scientific history—envisioned a platform approach to drug discovery, development, and delivery that moves quickly and scales. “How do we create a system by which we can always ride the wave of innovation? We can move with agility; we can pivot fearlessly and take on a new technology at the drop of a hat so that we can have a massive impact on technology,” he said.  

Risky Business

But pharmaceutical companies are inherently risk averse, Federer pointed out. “That's a good thing; we're working with people's lives. We can't move fast and break things,” she said. Rau countered that only some parts of the drug development process truly need to be risk averse.  

“It’s true that we are in a regulated industry that has speed bumps in it that makes you slow down, but the regulators are only in part of what we do, only really a fraction of what we do,” he said. “Certainly you want to have speed bumps and you want to make sure that you’re going slowly and safely when you’re conducting a clinical trial. The problem that I see is that we apply that mindset—that clinical trial mindset that you need to take 18 months or six years to do something and you need to do it absolutely perfectly—… to every single thing that we do. And in the world of technology, that’s a really bad idea.” 

Echoing Meyers’s initial point—that the drug discovery business is computational at heart—Rau argued for a technology development mindset: “Go really fast and get things done!” In fact, he added, if every single stakeholder agrees on a good idea, it’s probably so watered down that it’s actually a terrible idea. 

The panel shared their own examples of how moving fast within their organizations led to wins—and perhaps a bit of cultural upheaval.  

Rau mentioned LillyDirect, a program to connect patients directly to Lilly medicines and services. Initially the company outsourced development. “But we didn’t really go any faster. We didn’t end up with a better solution. It just looked really ugly and cost us a lot of money,” Rau said. An internal development team took over and in six weeks built a second version that was fast, scalable, and agile.  

Speed is innate at Moderna, Miller said. Within four weeks of the first externalization of large language models (LLM), Modern had built mChat, an internal LLM to automate document upload and scan internal libraries. Miller reports more than 2,000 active users daily and more than 5,500 total users. “It’s become the largest used capability in our company, but it didn’t come just by putting a technology out there,” he said. “I would argue 70% of the work comes from the culture transformation that has to go along with the technology.” 

At Bristol Myers Squibb, Meyers trained GPT4 on 5,000 clinical trial protocols and asked the algorithm to create new protocols for clinical trials. He was prepared for the backlash. “Of course, immediately everyone: ‘Oh, my God, there's just no way!’ And, ‘This is so high risk!’” he recalled. But the plan was never to use the AI-generated protocol as-is. Meyers estimated that the generated protocol was about 70% good enough, then “many, many humans-in-the-loop” stepped in to complete the work. And yet, he said, having a starting point shaved months off trial set up.  

Takeda’s patient-first approach has pushed the organization toward at-home technology not just for monitoring but for gathering information on disease biomarkers. “We're looking at at-home EEG,… both the technology aspects of that, and how do we enable that in a compliant way for our clinical trials,” Heatherington said. Takeda in bringing their AI/ML technologies into the patient’s home to gather data about sleep signatures, staging of sleep, the impact of drugs, and even diagnostics, she said. “All of that is to prevent these patients having to come in and be wired up in a sleep lab. It makes a dramatic difference in patients’ lives if we can do this.” 

In each use case, the technological hurdle ran in parallel with cultural hurdles, but Meyers noted that technology wins often make cultural converts.  

“For us [at Bristol Myers Squibb], I think it’s been really just getting people to open their minds, being willing to experiment and then really be impressed with some of the results they see,” Meyers said. “I think that creates a flywheel effect where people get inspired and motivated by what novel innovation and technology can do, and that encourages more people to want to experiment, and it just builds and builds and builds. And then all of a sudden, all the usual inertial forces in the organization don't have the grip that they had anymore.”  

Innovation at Scale

Moderna’s Miller argued that expediting wholesale change—either inside an organization or even across the industry—will require the convergence and standardization of data. “When something is standard—like a patient name is a patient name is a patient name, [or] temperature equals temperature—there’s less opportunity for errors in your system if you’re going to use that data through machine learning platforms,” he said. Data needs to be accessible and supported by a foundation of engineering excellence so that tools, applications, and cloud computing all work consistently and reproducibly.  

“I really believe in the saying, ‘If it doesn’t scale, it doesn’t matter!’” he said. “No matter how small [a tool], you don’t know the next use case. So you’ve got to build it in a way that’s decoupled, and that it can move forward very rapidly,” he said.  

Rau, with a background running Apple’s retail and online stores, has seen the requirements needed to achieve global compatibility and he pointed to both the payments industry and aviation for standards that allowed global scalability and ease of use for consumers. “The reason those things happened were because of industry forces coming together to actually change things,” he said. “Now, of course, regulators had their part in both of those to help make it, but it was really a push from industry. And I would like to think that if we do some pushing from this industry, we can actually shape and fix some of these problems that we have.” 

The needed scale for drug discovery is not insubstantial. For instance, to scale across Takeda, a tool or solution needs to work well at 200 different sites in 30 geographies, Heatherington said. “That sort of very fragmented ecosystem is in many ways its own worst enemy in terms of its uptake and the impact that we can have,” she said. But she’s hopeful: “The medtech industry has exploded recently…, and that’s critical for our ecosystem and what we do. I simply can’t wait for the day that that ecosystem matures sufficiently for us to be able to use it,” she said.  

She also pointed to the GDPR and other legislation on data privacy and data governance from Europe. Groups watching those discussions are developing guidances on trusted data partnerships, governance, and infrastructure that includes patient trust and confidence around how those data come together, she said, be it the electronic medical records, patient outcomes, or the other information made accessible to different stakeholders in that ecosystem. A crucial point, Heatherington added, is to be aware of technology ethics, to “really ensure that we’re doing the right thing for the patient as we apply all these technologies.”  

Miller closed the conversation with a challenge to disruption. “Be bold, be a change agent, be disruptive,” he said. Standardization is disruptive, he pointed out. “Differentiate yourself as a business by the impact you’re having on humanity, not by hoarding [data] in a closed loop system. The more open we all get, the more we’ll share data… We have to figure out how to remove the viscosity from that system so we can have the greatest impact on humanity.”