AI in drug development: The FDA's open-source shift and the scientific upside

Q: How does the FDA's adoption of open-source tools change what pharma companies must do?

The FDA now expects access to raw data and open-source programs so reviewers can execute their own analysis and compare it against the sponsor's results. This requires companies to transition away from siloed legacy tools and establish modern, compliant submission pipelines that natively support R and Python. Building these reproducible, governed environments ensures your data pipeline is structurally aligned with regulatory expectations.

Q: What is the practical difference between operational AI and scientific AI in clinical trials?

Operational AI functions as a workflow and coding assistant, accelerating routine tasks like generating standard data specifications, writing code, and shortening submission timelines. Scientific AI uses machine learning to integrate complex, multimodal datasets, such as genomics and imaging files, to uncover hidden disease pathways and predict patient behavior. While operational efficiency caps your cost savings at 100%, scientific innovation offers limitless upside for new therapies.

Q: How should pharma organizations navigate the SAS-to-R talent transition?

You must acknowledge that new data scientists entering the workforce are already R and Python practitioners who expect to use AI assistants out of the box. Instead of building basic training programs for new hires, companies must provide flexible, validated tech platforms that accommodate modern data science habits. For senior practitioners who spent their careers in SAS, the expectation is to actively engage with the open-source frameworks that are now standard across drug development.

FDA reviewers now expect raw datasets and R packages to run their own statistical validation models. This structural change requires pharma teams to overhaul their submission pipelines, shifting the true value of clinical AI from simple operational speed to scientific breakthroughs.

Eric Gibson

SVP, Global Head of Advanced Quantitative Sciences

Novartis

What you'll take away from this session

The FDA's pivot to R represents a permanent regulatory shift

FDA reviewers now want raw datasets and functional R packages to conduct independent, internal analyses rather than just reading static submission documents.

The real return on pharma AI is scientific innovation rather than operational speed

Machine learning can characterize complex disease mechanisms, isolate optimal patient subgroups, and design trial frameworks that were previously impossible to conceptualize.

Synthetic data allows engineering teams to front-load pipeline coding before a trial begins

Modeling highly accurate artificial patient data lets programmers write, test, and debug their CDISC-compliant dataset pipelines long before the first patient is enrolled.

New graduates arrive with open-source and AI skills natively embedded

Universities no longer prioritize teaching legacy tools like SAS. New hires enter the market requiring modern, governed platforms that match how they already code.

Real-time trial data evaluation unlocks advanced, adaptive study designs

Accumulating efficacy and safety data continuously allows sponsors to optimize dosing, re-estimate sample sizes mid-course, and significantly compress regulatory timelines.

Technology cannot compensate for weak trial design

AI can supplement data processing, but it will never replace a multidisciplinary team of scientists choosing the correct endpoints, comparators, and statistical methodologies upfront.

Eric Gibson, SVP of Advanced Quantitative Sciences, oversees 1,200 researchers at Novartis. Drawing on the massive scale of his organization, he tackles a central question facing the entire pharmaceutical industry: what does this current wave of clinical AI actually mean, and why does it feel different from past technological shifts?

The answer is divided into two parts. First is the operational reality. The regulatory landscape has permanently changed. FDA reviewers are highly trained quantitative scientists who want raw data and a suite of R packages to generate their own insights. Drug development teams must build open-source, reproducible pipelines to align with this regulatory direction.

Second is the scientific upside. Gibson highlights Novartis’s five-year collaboration with Oxford University on multiple sclerosis, where machine learning algorithms analyzed a multimodal dataset of clinical records, fluid biomarkers, and a quarter-million brain MRIs. The AI successfully isolated eight distinct disease states and identified the transitional biology driving disability progression. This discovery prompted regulatory meetings with the FDA and EMA to redefine how MS trials are designed. For pharma organizations targeting this level of innovation, establishing a governed statistical computing environment is the foundational step.

FAQ

How does the FDA's adoption of open-source tools change what pharma companies must do?

What is the practical difference between operational AI and scientific AI in clinical trials?

How should pharma organizations navigate the SAS-to-R talent transition?

Transform the work that matters most

See how Domino helps the world’s most regulated enterprises build, scale, and govern AI-powered applications.

Watch demo Talk to an expert