AI and the future of pharma with Bob Bradway, CEO of Amgen

When you can free up a scientist and you can take a scientist who was otherwise spending hours requisitioning materials or summarizing data or filling in forms and do that for them agentically, that's magic.

Thanks. Nice to be here with you.

Well, we were interested in human genetics as a way of selecting targets that might be relevant for human disease. Our notion was that if we could acquire a curated set of data over a large enough population of people, that we'd be able to draw insights from which we could base some of our discovery research efforts. And there were two things that were going on at that moment that were particularly attractive to us. There existed a company, deCODE, that had been formed some number of years before and had captured the genetic information on virtually everybody on the island of Iceland. So you had a high-quality longitudinal database linking genetics to physical and health traits, from which we thought, again, we'd be able to harvest insights that would point us to targets that would be more likely to be successful in the long run. The second thing was our imagination was captured a little bit by IBM's Watson, and we knew that at some point artificial intelligence would enable us to make sense of large data sets like that which we were assembling in Iceland. Now, it was still a little bit early then to talk about AI in the way we do today, but it was two things coming together: a belief that the sequencing technology would enable us to do it in vast numbers and at a time frame and at a cost that was reasonable, and that at some point AI would change the game as well. That was behind that decision.

Yeah, we did. In fact, I remember talking to Andrew about our genetic information in particular, and his response at the time was to say, 'Well, that's interesting, but it's not tractable today, and I'm interested in things that are tractable today, things that we can do today.' Whereas there were things in that moment, image processing for example, computer vision, that were well suited to the challenges of our business. So we began a dialogue with Andrew then about how we should think about machine learning and how we should think about artificial intelligence in the biotech industry.

That's right. And you were very clear about wanting to focus on the wonderful short-term bets.

Well, again, we knew that if we were talking about vast sums of data, that would require mathematics and technology processing capabilities that would be quite profound. We have, of course, through the years now, we've installed a super pod, an NVIDIA super pod in Iceland, and we use that technology to interrogate those data sets. But you're right that we knew that in making a commitment to that large treasure trove of data, that we were going to need to have the processing and the technology and the mathematical capability to make sense of it over time. But there's something else, Andrew, that may not be obvious from the outside, which is that AI scientists have been attracted to the area of protein folding for decades, catalyzed by a Nobel Prize winner who ushered a challenge of sorts, suggesting that it should be possible to predict how a protein will fold in a cell based simply on the sequence of amino acids of that protein. And in 2021, of course, DeepMind revealed what is now well known as AlphaFold and demonstrated success in being able to predict in silico what had previously been impossible to predict. Getting the crystal structure of a protein is a complicated journey. It requires very talented people, expensive technology, takes a lot of time, and you're not always successful. And so, the fact that DeepMind was able to crack that nut and publish it in 2021 created a rush of excitement in our field.

And we've since then, we've developed our own models. We continue to absorb and test other protein folding models and have moved beyond that to try now to actually design drugs, and for example, de novo antibodies is the current exciting topic in the field. But you're right that we're not relying on any single approach to this, but rather seeking to bring our expertise, our data, evolve the models that are available in ways that make them useful to us. In our business, we need to know the structure of a protein. So we need to know how it folds in order to gain insights into what its function might be or gain insights into what the problems in manufacturing it and delivering it to patients might be. But our interest is, of course, in designing medicines that are going to be effective in humans. So we've learned through decades of doing this that there are certain attributes of a molecule which make it either difficult to manufacture or create a risk of immunogenicity, and both of those can be liabilities which show up at the end of a 10-year, $2.5 billion journey. So if you can use technology and data and insight up front to make the changes necessary to improve your likelihood of success, that's obviously a big deal for us. So again, we were very fortunate that we had this notion that protein folding would be possible someday with the tools of artificial intelligence, and that if artificial intelligence did crack that nut, we would do some things differently. And we had begun assembling some of those things that we would want to do differently in anticipation of that moment. Now, in fairness, it came much more quickly and it was probably more profound even than what we had expected. And then, like everyone, I think we realized that AI was going to be a big deal, and we needed to figure out where it was going to be a big deal in our business and then invest around those areas.

Well, I think our business does start with innovation. So the core of our strategy is innovation. If we don't have an innovative new idea, an innovative new molecule, we really don't have a business. Now, so our initial effort in artificial intelligence was around the things that happened very early on that journey. So I mentioned protein folding a moment ago, but it starts at that early stage. And what we've tried to do is work downstream from that and say, okay, once we have a molecule, what are the other processes, what are the other things that we do in order to get it developed and get it registered and then make it available to patients and then support it in manufacturing and so forth. So we've tried to understand end to end the processes that we engage in and then to define or identify those that might be amenable to the tools of artificial intelligence. And then to say, okay, of those places where we could intervene, in which places should we intervene? And of course, that changes from one project to the next and it changes as new technologies have been introduced along the way. So, as an example, using AI today, we're able to select the molecules that we want to move into clinical development much more quickly. Let's say we're saving 50% of the time that it once took us to get a molecule, to make the choice of which of the molecules that we think will bind the relevant target is the right one to move into pre-clinical and then clinical experiments. So, we're moving more quickly there. So, then we say, okay, well, what's the next step in the process? And is there something that technology would enable us to accelerate? And so the process goes. And then right through the value chain, manufacturing in our environment includes a lot of visual inspection, and that of course is an area that's been ripe for AI intervention for some time. And so there, and capacity and modeling and so forth, that over time we'll do more of with AI. And then sales and marketing. Obviously, we're excited about what might be available down the road in the way of sales and marketing. There are some AI tools at the moment which clearly are changing how our customers consume information about either our medicines or the diseases our medicines treat. So literally each of the different stages of the way in which we engage in business, we have groups trying to figure out are there things we could do differently with AI. And again, if we could do it differently, should we do it differently.

Well, I think part of what we do is we brainstorm and we try to think for the future and say, not now, but what's possible five or 10 years down the road. What do we see coming that may have a big impact on our business? So a current example would be quantum. I think by now we all recognize at some point there will be quantum computing, and quantum will enable us to tackle some challenges that are difficult to break down with classical computing. And in our field, there are some specific areas where that's true. You mentioned the three-dimensional and the movement of proteins. For example, some of the movement and the way in which proteins fold and then react in different environments may be more possible to model with quantum than is feasible with classical computing. So we have begun with respect to quantum to begin to identify the quantum equations that are relevant for our business, the quantum experiments or the quantum technology solutions that may be possible, and begun to think, okay, if it were available, what would we do with it? So we try to think about strategy in that fashion to say, what are the big changes that might be coming our way? How would we want to get in front of those big changes to harness them for the benefit of what we try to do, which is discover, develop, and advance medicines that make a big difference for patients that are struggling with serious disease. And so in that way, we're not trying to say it will happen by a certain date and we need to be ready, but rather if it happens, what would we do? And then try to line up some capabilities so we're not surprised and we're ready to move quickly.

Yeah, it's interesting that you frame it around change management. I think that's right. I think this is a huge change management process that we're all at the early stages of.

I hope you're right, because I think being late to this party is going to be expensive. I think being early to the party has the potential for real benefit. So I hope you're right that we're ahead of some of the competition. But for different reasons unrelated to AI, we needed to make some profound changes to our organization in 2013. And we decided at the time that the high fruit of that would be learning how to lead change management in the organization. And Amgen had been such a successful company for so long, we had had decades of success and for many years uninterrupted success. But the patents on our first sets of molecules approached the date of expiry. You know, in our business when a patent expires, revenues plummet and profits plummet with it. And as our first set of molecules were approaching that point, we realized we needed to operate differently. And so we focused not just on what had to change for the benefit of the loss of those patents, but we focused on how would we lead company-wide change processes. How could we teach the organization and learn ourselves as leaders what's required to get lots of people to move together and make change. And that's hard, but fortunately we created some leaders along the way who know how to do it and do it well. And those same leaders are being useful in the context of AI. So for example, Dave Reese, who has served as our chief technology officer for the past couple years, was one of the instrumental leaders in helping us make the changes that I referred to beginning in 2013, leading in particular changes that we needed to make in research and development and at the intersection of research and development with process development and business development and so forth. So leaders like Dave cut some important teeth on change management many years before this AI moment arrived and can leverage that skill now.

Well, I don't know about non-obvious, but I think if you're not leading it from the top, I think it's easy for people to ignore it. You know, I think if the CEO is enthusiastic about AI, after a while, people will start to figure out whether they're enthusiastic, too. But I think at Amgen, for better or worse, people know that the senior leadership team believes that the suite of tools that fall under the umbrella term artificial intelligence are a big deal and are changing the way we compete and the way we operate and the way we run the business.