Convergence: ‘Get ready’ to use genAI in drug development

PITTSBURGH — Use of generative artificial intelligence (AI) in drug development is no longer theoretical—the technology is here and being used by sponsors and regulators in drug development, experts told attendees at the RAPS Convergence 2025 conference.

To date, there have been more than 100 drug submissions to the US Food and Drug Administration (FDA) with AI and machine learning components in them, Rudy Fuentes, PhD, RAC director, regulatory affairs at Thermo Fisher Scientific, said in his presentation at the meeting.

“That just tells you where this technology is,” Fuentes said. “It’s not experimental anymore. It’s coming to life, so we’ve got to get ready.”

Within the last year, FDA launched two initiatives in AI: a draft guidance document published in January that proposed a risk-based credibility framework for use of AI in product submissions that would support regulatory decision-making, and change in its requirement for testing of monoclonal antibodies and other drugs to allow for new approach methodologies (NAMs) such as AI-based toxicity and cell line models as well as organ-on-a-chip systems. (RELATED: AI in drug development: FDA draft guidance addresses product lifecycle, risk, Regulatory Focus 07 January 2025; RELATED: FDA seeks to reduce animal testing requirements for mAbs, other drugs, Regulatory Focus 11 April 2025)

Sponsors can integrate AI into regulatory tools like the Regulatory Information Management System (RIMS), Regulatory Intelligence Hub, Publishing Solution, and Clinical Trials Suite of Systems (CTMS), Fuentes said. It can also be used to reduce the amount of time spent and problems associated with an investigational new drug (IND) application, he noted. For instance, a generative AI tool can have structured inputs for RIMS or another platform, generate an initial draft, and have it reviewed by a human. The AI tool can then formats, validate the structure of, and package the submission for FDA, he said.

Fuentes acknowledged that implementation of generative AI currently has several limitations, including data quality and reliability, technical limitations, a lack of experts across related industries, and a lack of standards.

“It's a promising technology, but we have to be careful what we do with it,” he said.

EMA and generative AI

Florian Lasch, PhD, a biostatistics specialist with the European Medicines Agency (EMA) told attendees generative AI is used in regulatory submissions, process analytics to improve efficiency, and healthcare analytics to structure information.

“We are both regulator of AI and users of AI, and I think that's a unique position,” he said.

Lasch said that EMA “recognizes the transformational potential of AI and the positive use cases” while also acknowledging that “there are risks that need to be managed,” which include patient risks and regulatory risks.

EMA has supported use of AI in medicine through publishing a reflection paper on the use of AI in the medical product lifecycle and will publish guidelines on AI in clinical development and pharmacovigilance within the next year, Lasch said.

The agency is also working with regulators across the globe to “to progress the capacity building within the network and to ensure global harmonization where possible as early as possible,” he added.

Generative AI use cases

Zach Weingarden, MSE director, AI technology and applications at TrialAssure, said implementing AI into medical writing workflows can allow for faster drafting, fewer quality control loops, a consistent voice across a clinical study report (CSR), plain language summary, and synopsis sections. AI can also serve as a guardrail for compliance by identifying non-compliant claims, he noted.

Sponsors should have a way to identify which components of a document were written by an AI and reviewed by a human, Weingarden said.

Xinjiang Wu, PhD, associate director, medical writing at Biogen, said that his company is using generative AI to assist their medical writers with drafting CSRs, but intends for the technology to eventually expand to help with more complex documents such as briefing books and module 2 summaries.

He said that AI should serve to empower medical writers, not act as a replacement for them. Medical writers at his company “essentially still retain ownership of the document,” he said.

“Strong communication support and clear role definition is very important. It is crucial for the AI adoption in the organization,” Wu said.

In the future, Wu said his company hopes to embed AI into standard operating procedures, workflows, and regulatory timelines as well as accelerate the timeline of submissions, and build cross-functional alignment across regulatory authorities and other stakeholders.

AI implementation can be challenging. “There's not so much a standard approach yet to this, he said. “It's going to depend on what systems you use and what workflows and how we're going to implement AI into your processes,” Weingarden said.

Where there are challenges, there are also opportunities, Weingarden noted.

“I think everyone kind of recognizes that it can be very powerful, but it also has to be done in a way that's responsible and aligns with in terms of compliance and regulatory best practices as they evolve, whether we're talking in the pharmaceutical industry or anything else, really,” Weingarden said.

RAPS Convergence

Convergence: ‘Get ready’ to use genAI in drug development

Related topics

Recommended content

Welcome to the new RAPS Digital Experience