.png)
Drug development in biotechnology takes time. A lot of time. On average, it takes 10 to 15 years to take a medicine from initial discovery to approval for patient use. For patients, that means waiting—sometimes too long. For scientists, it means navigating a maze of trial-and-error.
Amgen, a Southern California-based global leader in biologic medicines fighting some of the world’s toughest diseases—including cancer, cardiovascular disease, inflammatory diseases and rare diseases—is using new technologies to help transform its R&D process. The company is weaving artificial intelligence (AI) and machine learning (ML) deep into the scientific fabric of drug discovery to potentially accelerate the development of medicines for seriously ill patients.
“We’re using AI to help pose new hypotheses and design new molecules,” says Howard Chang, Amgen’s chief scientific officer and head of Global Research. “This leads to deeper insights—and faster innovation that may open paths to therapeutic solutions previously thought to be out of reach.”
Where AI Sparks Ideas
Human genetic data on a massive scale make this innovation possible, since genetically validated targets have a greater likelihood to result in successful medicines. Scientists working at the company’s site in Reykjavik, Iceland, have access to one of the richest human genetic datasets in the world, including millions of genotypes, hundreds of thousands of whole-genome sequences, reams of transcriptomic and proteomic profiles, and thousands of phenotypic traits.
“Amgen has long been rooted in this idea that human genetics will lead to better medicines.” Chang says. “Investments in diverse population datasets and supercomputing power are helping our scientists to translate insights into medicines. We’re able to innovate with confidence and deliver with urgency while staying grounded in human biology.”
Amgen's AI/ML approach also has guardrails to reinforce governance and controls and safeguard the security, privacy and protection of data.
Designing and Optimizing Drug Candidates
Today, nearly every molecule in Amgen's early research pipeline is shaped by an AI-driven approach. These approaches have accelerated the speed of protein engineering, the complex, structural work of designing a potential medicine.
Once a target is identified through genetic insights, generative AI tools help scientists to engineer novel proteins for drug-like behavior before they ever touch a test tube.
Predictive ML models take it a step further, forecasting how those proteins will act in the body. One Amgen-built model can predict a protein’s viscosity, a key metric for injectable drugs, with more than 80 percent accuracy using only its amino acid sequence.
“What makes Amgen’s approach unique is its tight integration between AI models and the rest of drug discovery and development,” Chang says. “We’re not just using off-the-shelf tools—we're developing fit-for-purpose systems specifically designed for therapeutic protein engineering.”
This system-focused approach means Amgen’s data collection and model development are deeply aligned with the next steps of creating a new medicine, including determining manufacturing success rates, clinical performance, and, ultimately, whether a drug works for patients.
AI Tools Built for Medicine
This kind of acceleration in R&D wouldn’t be possible without ML tools and AI systems designed for biology. AMPLIFY, an open-source protein language model developed in collaboration with Mila AI institute in Quebec, Canada, learns the “grammar” of proteins—mapping amino acid sequences to structure and function. It makes sense of the “sentences” within the genome so scientists can design better proteins from the start.
EVOLVEpro, developed with contributions from Amgen Science Fellow Kaiyi Jiang, takes it a step further. “Most protein design models today are trained on public sequencing data, essentially natural evolution, but evolution doesn’t select for things like avoiding cancer. That's where medicine comes in,” Jiang says. “EVOLVEpro generates the data that allows us to build on evolution to find better ways to treat complex diseases.”
Rather than simply analyzing proteins, EVOLVEPro proposes smart mutations, learns from the outcomes, and then uses that experimental data to refine its predictions—creating a self-reinforcing loop. In published tests, EVOLVEPro boosted protein activity up to 100-fold across antibodies, CRISPR enzymes and RNA polymerases—all within just a few experimental rounds—demonstrating how each cycle strengthens the model, which then drives iterating experiments. For the first time, AI is teaching scientists to do experiments that will improve AI models.
Together, the protein language models form a powerful pair: one uncovers the patterns within the genome while the other applies them to design better proteins, faster.
Beyond the Algorithm
Looking ahead, the developers and scientists at Amgen are working to scale AI/ML technologies across their entire drug pipeline—from target discovery to regulatory filings. These tools can also be used to find how existing medications can work better for more patients. Ultimately, the goal is to find treatments for the 85 percent of disease targets that have eluded medicine up to this point.
That’s a big leap. But Chang believes Amgen is not just chasing speed or efficiency. It’s hoping to change what's possible.
“At the end of the day, our mission to serve patients remains constant,” he says. “This is the next step in how we bring biology and technology together to create innovative medicines that help fight the world’s toughest diseases.”
