PCSK9 inhibitor (AstraZeneca)

THOUSAND OAKS, Calif. — In 2023, Dave Reese walked out of a Silicon Valley meeting and made a career-changing call. Reese was working as the head of R&D at Amgen. He rang up his CEO Bob Bradway and, effectively, pitched the idea of firing himself, giving a simple explanation: Everything in biotech was about to change, and Reese wanted to lead that transformation for Amgen. Over 20 years at Amgen, Reese had long envisioned a convergence of biotech and tech. The AI boom accelerated his belief that the time had come, spurring him to pitch the new role of chief technology officer. In doing so, he’s reorganized large chunks of the 28,000-employee biopharma to capitalize. “It’s the hinge moment,” Reese recalled telling Bradway. “There’s no question.” Last month, Reese and other Amgen scientific leaders gave Endpoints News an exclusive look at how the AI strategy is playing out. The reality, Reese says, is there will be no Eureka moment that changes everything. Instead, Amgen’s labs are undergoing a continuous, radical transformation, putting AI models at the heart of more and more of the process of finding and developing drugs. Those efforts, they say, are starting to deliver. Computation in drug R&D isn’t new. For decades, drug developers have simulated and generated molecules on the computer with technologies named like QSAR or molecular dynamics. But, in Reese’s view, AI is a “step change” difference, similar to how the telegraph was getting incrementally better until the telephone arrived and “obviated the need for that.” “We’re at a sort of telephone moment,” Reese said in an interview at Thousand Oaks, dressed in a gray blazer and blue-striped shirt on an unseasonably hot September day. And while most of today’s gains are making things faster and cheaper, Reese said that alone is not the ultimate goal. “The efficiency play is important, but that’s not the prize,” Reese said. “The prize is getting undruggable targets even earlier in the process. The prize is understanding disease biology that we don’t yet understand, where we don’t even have a target or we don’t have tractable ways of going at a disease.” Yaron Werber, a TD Cowen analyst who started covering Amgen in 2006, credited Reese’s decision to leave the R&D head position as a key boost to the changes of making AI work inside Amgen. Reese brings as deep a familiarity as imaginable in what pharma R&D is about. “That’s fairly unique, because AI usually is driven by informatics people and computer scientists who don’t really speak biology,” Werber said in an interview. “There’s usually a bit of a cultural clash, potentially, so developing that in partnership, organically, is very powerful.” A lab inside the uncreatively named Building 29 is home to a research team embodying a lot of Reese’s desired changes. At the center of protein processing platform lab — known inside Amgen as the P3 Lab — is a robotic arm that whizzes around, grasping a plate and moving it to one of six plexiglass-covered workstations arranged in a surrounding hexagon. “This is a future version of me,” Marissa Mock, director of generative biology at Amgen, said, staring at the robot. “Marissa 2.0,” she called it. In the early 2010s at Amgen, Mock did much of the robot’s work by hand, running tests to observe various qualities and activities of proteins. There’s nothing extraordinarily special about the robotic arm itself, which has become a commonplace feature in any modern lab. When Endpoints visited, the lab scientists booted up a demo that puts it in a frenzied, near-constant state of motion. When reporters aren’t watching, it typically moves every half hour or so, moving plates full of proteins onto the next step in whatever lab test it’s running at the moment. The lab’s main charge is turning initial hit candidates into more optimized drug programs that are ready for the next preclinical stages. The robotized system runs over two dozen types of lab tests in characterizing these proteins, studying their qualities and activities. Viscosity is one of those characteristics. Too thick, and it can’t be pushed through a syringe. Too thin, and it may require more injections to get enough of the protein delivered as a drug. Traditionally, Amgen’s scientists would manufacture a few hundred milligrams of any protein of interest to assess viscosity. Today, Amgen runs an algorithm predicting viscosity from the sequence of a protein, before it’s ever made in the lab. If the prediction is not promising, it’s not made. “We now do it every day on every molecule,” said Alan Russell, head of R&D technology and innovation at Amgen. “There are literally 20 or 30 other predictive models that we use to correct all of the therapeutic molecular attributes that could cause a problem.” Russell, formerly a Carnegie Mellon University professor, joined Amgen five years ago and has seen every aspect of large-molecule discovery change over that timespan. AI models are playing a leading role. “We’re not at a tipping point,” Russell said of AI’s impact. “We’ve tipped. We’ve tipped years ago.” All the proteins made in the P3 Lab first go through AI models from computational biologists Marti Head and Chris Langmead, who work closely with the P3 Lab after both joining Amgen in 2022. Mock recalled that the previous digital chemistry team operated differently: more aloof and acting “almost like service providers,” she said. Now, the teams frequently meet and share a language spanning life and computer sciences. The computational team often has batches of proteins made, where they expect some will be terrible performers. But models benefit from negative results too, not just successes. Human scientists still make the final calls on which proteins are made. But AI models — like the viscosity-predicting one — are informing those decisions. Amgen hasn’t published or shared specifics on most of these internal models, but the strategy shows the next frontiers for AI in biology and drug hunting that extend well past the initial triumph of Google DeepMind’s AlphaFold in predicting protein structures. “If you map 15 to 20 of these, you start to knock them down in in silico design,” Reese said. “All of a sudden, you’re in a new world. It’s not a big bang moment, but it’s the accretion of these over time, but then you look back and say, ‘This is completely different.’” Amgen is already an outlier by its sheer existence as an independent, self-sustaining biotech. It started in Thousand Oaks, roughly an hour’s drive northwest of downtown Los Angeles, over four decades ago. At the time, it went by the name Applied Molecular Genetics and had a handful of scientists in a 3,000-square-foot space wedged between an evangelical church and a Rolls-Royce detailer. Since then, Amgen has built dozens of buildings on roughly 100 acres here, ranging from R&D to sales to manufacturing. It is one of three drugmakers on the Dow Jones Industrial Average, alongside Merck and Johnson & Johnson. But despite being one of the original big biotechs, it also may not be the first place that comes to mind when someone thinks of the AI revolution in the industry. Startups like Isomorphic Labs and Xaira Therapeutics have raised hundreds of millions to advance Nobel Prize-winning AI breakthroughs in biology. Roche’s Genentech has put Aviv Regev, a computational biologist and a leader of the lab-in-the-loop strategy , in charge of early R&D. Some of these leaders, like Isomorphic’s Demis Hassabis, are even trying to build their companies without the traditional wet labs — an even more radical bet on AI’s future capabilities than Amgen is currently making. Many know Amgen for its feats in the courtroom and around the deal-negotiating table. There’s a sticky — if unfair — description that has followed the company around: a law firm with a biotech business on the side. Werber called it a “grossly outdated” viewpoint. Still, the reputation comes from Amgen’s lawyers being a formidable crew that has argued before the Supreme Court, battled the IRS and FTC in court, and has defended its biggest moneymakers from generic threats over the years. And despite its biotech origins, Amgen has often turned to pharma’s playbook of multibillion-dollar acquisitions to find its next legs of growth. It bought Onyx Pharmaceuticals for $10.4 billion in 2013, the psoriasis drug Otezla for $13.4 billion in 2019, and Horizon Pharmaceuticals for $27.8 billion in 2023. Those three deals, alone, are roughly equal to Amgen’s R&D spending over the entire last decade. All of that can leave its internal science overlooked. Its labs, though, have developed major first-in-class, or nearly first-in-class, medicines, including the PCSK9 inhibitor Repatha in 2015, the CGRP migraine drug Aimovig in 2018, the first KRAS-targeting cancer therapy Lumakras in 2021, and a DLL3-targeted bispecific cancer drug Imdelltra in 2024. Amgen has also maintained its investment in deCODE Genetics, a once-bankrupt Icelandic startup it bought for $415 million back in 2012. DeCODE specializes in sequencing entire human genomes, rather than just the protein-making parts of the genetic code. Amgen has now built an on-site Nvidia supercomputer in Iceland to find patterns hidden in that data. Efforts like deCODE convinced Howard Chang to leave a longtime spot on Stanford’s faculty and become Amgen’s chief scientific officer earlier this year. Chang, a leader in genetics research, envisions a future of “smart medicines with programmable outcomes.” In an interview, he described these medicines almost like automobile designs, with a combination of modular and boutique parts being used to build a drug. MariTide, which has captivated investors for the past year, is putting these ideas to the test. The drug is a bispecific peptide-antibody conjugate that both blocks a hormone called GIP while activating GLP-1. Other drugmakers are testing candidates that turn on GIP, instead. Cowen’s Werber said the conventional thinking has been that GIP antagonism could weaken bones. The deCODE database suggested bone health isn’t an issue , Werber said, and the early data looks to support that viewpoint. “Amgen bucked the conventional wisdom based on the deCODE genetics data in hundreds of thousands of women, and so far, that database has been proven to be correct,” Werber said. Cowen’s Werber called that innovation record impressive. But he also said Amgen has limited itself by capping its R&D spending over the years to boost profitability and margins. The “artificial cap really hurt them,” he said, especially in maintaining early leads in competitive, crowded areas like cancer. But Reese’s AI push is part of a new surge in R&D spending underway at Amgen. R&D expenses are forecasted to exceed $7 billion in 2025, an all-time high in Amgen’s history and a 20% increase from last year, which was already a 25% jump from 2023. Large-scale trials for its obesity drug MariTide and heart treatment olpasiran are key drivers of the budget, but Bradway also called out spending on technological advancements like AI, as well, on its most recent earnings call. Those bets will need to deliver, as the drugmaker and its $158 billion market value face one of the industry’s most severe patent cliffs. A Morgan Stanley analysis found two-thirds of today’s revenues at Amgen are potentially at risk from patent expiries by the end of 2030. Supercomputers and robot-powered labs cost money. Reinvention has to deliver tangible results in replacing those sales, while overcoming the challenges of bureaucracy and cultural inertia found in every large corporation. So, is AI just the latest shiny object that large drugmakers are trying to latch onto as a savior? Or is this technology actually starting to deliver the revolutionary impacts that its backers have long promised? Reese is searching for answers in the research lab and beyond. Instead of running hundreds of pilot projects “that seem really cool but never scale and never go anywhere,” he said he’s focused Amgen on a few big AI bets in areas like molecular engineering, clinical trials and manufacturing. He’s equally eager to talk about AI built into its new Ohio manufacturing plant, where cameras and algorithms now inspect lines of drug product instead of humans, who would take 30 to 40 minutes to clear each and every line. That doesn’t lead to flashy predictions about AI curing disease, but it does tackle the real-world inefficiencies that have long weighed on pharma’s productivity. “That half-hour savings translates into many, many days of additional production over the course of the year,” Reese said. “It’s not sexy. It’s never going to get a headline in the New York Times . ” But unsexy stories of viscosity predictions, manufacturing inspections, and faster trial enrollment add up to what Reese has bet his career on: a future of drug companies looking like hybrids of technology and data companies with biotechs. “You then have brought the hinge moment to life at the enterprise level,” he said. “That’s the end goal, and I think it’s now inevitable. The only question is who’s going to get there.” And as with any new technology riding the hype cycle, there are some early reality checks on generative AI’s impact on business. A recent MIT study drew headlines by finding 95% of AI projects didn’t deliver businesses a return on investment. Reese said he pays no mind to such reports. “You’re running a marathon, and you’re a quarter-mile in, and someone’s declaring the outcome of the race,” Reese said of the MIT study, adding he pays “absolutely no interest” to these types of studies and resulting articles. “It’s just silly.” And then Reese leaned forward, across the table with a widening grin. “I hope our competitors are deeply skeptical,” he said.