The integration of artificial intelligence into biopharmaceutical research and development is advancing, with the U.S. Food and Drug Administration announcing on Apr. 10 the use of Elsa, a generative AI tool, to support parts of the drug approval process. While this regulatory adoption marks an important step, experts say it is only one aspect of a broader shift within the industry.
The growing reliance on AI in guiding scientific, safety, and development decisions raises questions about whether companies are generating and structuring the right data to ensure these tools lead to safer and more effective drugs for patients. The quality of data used by AI models is crucial for their effectiveness.
"AI is truly only as good as the data we provide to it," said an expert from Charles River Laboratories. They emphasized that understanding how AI interprets data is key to ensuring its usefulness rather than becoming a misleading influence with real implications for therapeutic development and human risk assessment.
As drugs progress through research and regulatory processes, any mistakes or gaps in data can be compounded by AI systems that lack human intuition. Early attention to clean and organized data—including robust metadata—can make future analysis more efficient. Metadata helps contextualize experiments so that generated information remains useful over time.
Formatting diverse datasets into a common language requires collaboration between experimentalists and data scientists throughout all stages of research. Data harmonization remains a significant challenge due to the lack of formal industry standards such as FAIR (findable, accessible, interoperable, reusable) principles for drug discovery information.
AI-based tools have already transformed areas like small molecule design and protein structure prediction but still face challenges modeling complex biological processes or predicting toxicities. New approach methodologies (NAMs), including nonanimal models paired with AI algorithms, are gaining support from regulators like the FDA but require further validation before widespread adoption by developers.
Experts suggest that greater collaboration among contract research organizations (CROs), regulators, health authorities, and industry could help democratize access to noncompetitive datasets needed for training advanced models supporting NAMs. A trusted organization acting as a safe harbor could facilitate this transition toward viable alternatives while speeding up drug development timelines.
Looking ahead, experts anticipate that improved alignment around generating high-quality structured data will allow companies not only to accelerate best-in-class programs but also increase options when developing first-in-class drugs—and eventually enhance predictions regarding human safety.
