How AI Drug Manufacturing Is Changing the Game

How Drug Companies Accelerate Drug Manufacturing Using AI

Using AI, Pfizer is able to detect anomalies and suggest real-time steps for its operators as it aims to boost product yield by 10% and cycle time by 25%, Pfizer Chairman and CEO Albert Bourla said in the company’s 2023 annual review.

The pharmaceutical company rolled out its generative AI platform in 2023. “AI-powered manufacturing processes are increasing throughput by 20%, enabling us to deliver more medicines to patients faster,” Bourla said in the report.

Working with AWS allowed Pfizer to speed up development and distribution of the COVID-19 vaccine and manufacture the vaccine in 269 days instead of the usual 8-10 years, according to Lidia Fonseca, Pfizer’s chief digital and technology officer.

At the AWS Summit in Los Angeles on Nov. 22, 2024, Fonseca noted that Pfizer’s mRNA prediction algorithm delivered 20,000 more vaccine doses per batch. Pfizer’s internal generative AI platform, Vox, on AWS cloud services allowed the pharmaceutical company to access large language models on Amazon Bedrock and SageMaker.

READ MORE: What does the growth of generative AI mean for drug discovery and clinical trials?

“In manufacturing, Bedrock takes the optimal process parameters to identify what we call the golden batch and uses generative AI to detect anomalies and recommend actions to our operators in real time,” Fonseca says.

She adds that by using AI, Pfizer can search and collate data and scientific content in a fraction of the time.

“And algorithms generate and validate potential targets to improve our scientific success,” Fonseca says.

Moderna also used AI to speed up development of its COVID-19 vaccine. It deployed AWS Internet of Things, AI/ML and data analytics services to a connected environment incorporating intelligent biopharmaceutical manufacturing and supply chain processes, according to AWS. AI algorithms also allowed Moderna to automate quality control analyses and reduce hours spent on manual review aimed at improving production processes and logistics, AWS notes in a case study.

Novartis uses ML to develop smart manufacturing processes. Merck’s Manufacturing and Analytics Intelligence is an AI-powered platform on AWS designed to optimize its drug manufacturing processes, according to Sheeran.

AI in Pharmaceutical and Life Sciences 

In October, the UCSF School of Pharmacy received federal funding as part of the Advanced Research Projects Agency for Health initiative to accelerate drug development using AI. Biotech companies can use the open-source data sets and models developed as part of the project by the nonprofit Open Molecular Software Foundation and John Chodera, a computational chemist at Memorial Sloan Kettering Cancer Center.

UCSF plans to use AI to map the terrain of molecules that are unwanted or act in dangerous ways. By speeding up drug development and lowering costs, researchers can navigate around issues that occur later in the development process. Researchers are using ML to predict how molecules interact with anti-targets. 

“When you’re designing new molecules, you need to be able to predict the molecule’s properties, such as how long it will stay in the bloodstream or whether it will get chewed up by metabolic enzymes in the liver, and right now, those predictions are good, but not great,” explains James Fraser, chair of the Department of Bioengineering and Therapeutic Sciences in the UCSF schools of medicine and pharmacy. “And so, the hope is that new advances in machine learning and artificial intelligence, when fed the right data, which we hope to generate, will tremendously increase the accuracy of those predictions, enabling us to synthesize fewer molecules to get to the same place, thereby speeding up drug discovery and making it cheaper.”

Drug creation company Absci uses AMD’s Instinct accelerators and ROCm software to power AI drug discovery workloads, such as Absci’s next-generation antibody therapeutics. AMD says Instinct GPU accelerators and ROCm software enable high-performance computing as part of an open ecosystem. On Jan. 8, 2025, Absci announced it would receive a $20 million investment from AMD to advance this research and meet the demand for AI applications in drug discovery.

“One of the things we’ve focused on are what we call undruggable targets,” says Sean McClain, founder and CEO of Absci. “By being able to drug a target, you can modify the pathway affecting the underlying disease, creating a potential cure or potential treatment.”

Absci uses generative AI models to design antibodies that bind to cancer targets, modify the pathways and kill the cancer, according to McClain. He says AI has helped speed up the time it takes drugs to reach clinical trials from five and a half years to 18 to 24 months. Absci has developed an antibody on its AI platform for inflammatory bowel disease.

He also says drug companies can use AI models to help when filing investigational new drug applications with the Food and Drug Administration to get approval to test drugs on humans.

“Looking forward, there are still a lot of gaps that AI hasn’t been able to solve yet that I think over time it will, but I think it’s already making a dramatic difference in how we go about designing and creating drugs,” McClain says.

RELATED: How can the cloud improve data sets for real-world evidence in clinical trials?

What Drug Manufacturers Should Consider When Using AI

When deploying AI in drug manufacturing, organizations should ensure they have the data infrastructure to collect, store and analyze the large data sets that AI requires, Sheeran advises. He adds that organizations should have a clear strategy on how to integrate AI into the drug manufacturing workflow and validate AI.

“Companies should also prioritize transparency and explainability in their AI systems,” Sheeran says. “At AWS, we work backward from our customers’ needs and desired business outcomes to help them navigate these considerations and implement AI solutions responsibly and effectively.”