How Agentic AI Identifies Toxicology Data Gaps and Testing Options

Ahead of the American Association of Pharmaceutical Scientists’s (AAPS) 2026 PharmSci 360 conference, Thomas Luechtefeld, PhD, founder and CEO of Insilica, spoke to BioPharm International®. In the second segment of his interview, he discusses why evaluating the process behind artificial intelligence (AI) decisions can matter more than accuracy. Dr Luechtefeld is scheduled to present “From Read-Across to Generative AI: The Startup Stack for Animal-Free Toxicology” at the conference during the Hot Topic session “The Countdown to Post-Animal Drug Development: Disruptors on the Ground” on Oct. 28, 2026.

Dr Luechtefeld posits that benchmark accuracy alone is not enough to trust an AI system's toxicology decisions. A model that scores well can still be biased, he explains, and many large language models are poor at judging their own confidence.

"I would even argue that being able to evaluate the process behind a decision is more important than the accuracy of the decision itself, because over time we can refine our systems to improve the accuracy," Dr Luechtefeld states.

Understanding the decision context makes decisions reproducible and improvable, he adds. If an agent reaches a conclusion without supporting information, a person, another agent, or a deterministic system can audit that gap and judge the decision's reliability. Recognizing what is unknown becomes critical as AI makes more decisions.

How can AI systems help identify toxicology data gaps?

Dr Luechtefeld notes that agentic systems can now identify data gaps, a task previously too labor-intensive. A system can determine which experiments exist in public databases for a given chemical and which have not been performed. It can then recommend experiments, identify vendors that can run them, and propose the most cost-efficient way to increase confidence in a decision.

Combined with new approach methodologies and a growing range of higher-resolution experiments, these capabilities signal a golden age of science, Dr Luechtefeld emphasizes. He expects the field to produce more, better, and safer drugs faster.

Learn more from the first segment of Dr Luechtefeld’s interview.

AAPS 2026 PharmSci 360 will be held October 25-28 in New Orleans. Attendees can visit Insilica at Booth 919.

Click here to explore more conference coverage.

About the speaker

Thomas Luechtefeld, PhD, Founder and CEO, Insilica

In addition to his role at Insilica, Dr Luechtefeld, is also the creator of ToxIndex, an AI platform for chemical and product safety assessment. ToxIndex coordinates specialized agents across thousands of databases, more than 100 million regulatory and scientific documents, and hundreds of predictive models. The platform assembles this evidence into structured risk assessments. Each claim retains its source, version, and audit history, allowing an assessor to inspect the evidence behind a conclusion and evaluate its relevance to a specific context of use. Dr Luechtefeld holds a PhD in Environmental Health Engineering from the Johns Hopkins Bloomberg School of Public Health. His research on chemical safety data, predictive toxicology, and integrated testing strategies has appeared in Toxicological Sciences, ALTEX, and Regulatory Toxicology and Pharmacology. He also contributes to the European Union Horizon ONTOX project and the National Science Foundation-funded BioBricks open knowledge graph for chemical safety.