Preclinical studies represent one of the most expensive and ethically charged phases of drug development, where a molecule's fate is decided before it ever reaches a patient. Dr. Steven Bulera, global head of toxicology at Charles River Laboratories, now divides his time between traditional oversight and a fast-moving experiment: folding AI into the workflows that generate safety data for regulators.
His assessment is clear-eyed. AI is delivering immediate value on the operational side-scheduling resources across 15 in vivo sites and accelerating report production. The scientific side, where algorithms would predict on-target and off-target effects from a molecule's structure, remains "a dream," he said. "A cell, or an organism, is a very complex system, and we don't have all the knowledge yet."
Administrative efficiency and AI-assisted reporting
Bulera breaks his AI work into two buckets: administration and science. The former is where progress is most tangible. His team is exploring tools that optimize resource allocation globally and slash the time needed to produce study reports. Even here, the output isn't handed off untouched. "Subject matter experts still need to review the material and make sure the data is interpreted appropriately," he said.
To avoid triggering client anxiety, the organization often drops the term "AI" in favor of "automation." Some pharmaceutical contracts explicitly forbid AI use or data entry into AI systems. "But to write a report, we need data," Bulera said. The answer is building tools behind secure firewalls, with information security teams vetting every step. Clients, he said, become receptive once they see the tool can halve report turnaround times.
Scientific hurdles and data limitations
Predictive toxicology via AI faces a pair of steep obstacles. The first is the sheer complexity of biological systems: signal transduction cascades and receptor interactions still hold too many unknowns. The second is data. Proprietary concerns keep pharmaceutical companies from pooling information. "On that side, nobody is pulling all of the data together. How can they, without losing their competitive advantage?" Bulera said.
He draws a direct line from this fragmentation to scientific stasis. Without access to failures-compounds that never reached the market-any training dataset is skewed. "We want everything so we can create a better data set to mine or a tool that's more predictive," he said. For now, AI can't replace animal or in vitro testing, and he doesn't expect that to change within a few years.
Regulatory acceptance: a 'wild west'
Global regulators hold no unified position on AI-supported safety studies. Some welcome the efficiency; others question whether an expert truly reads an AI-assisted report. "That's why we talk about AI-assisted interpretations-not just AI interpretations," Bulera said. The result is a fragmented landscape. "I can list an entire alphabet of regulatory agencies, each of which may have different opinions. … It's kind of the wild west."
Contract research organizations like Charles River might serve as neutral conveners, he believes. By aligning several large pharmaceutical companies around a common approach, they can approach regulators with a unified proposal and then iterate based on feedback. "From our findings, we can go to regulators, saying, 'Look, we have the support of these global pharmaceutical companies, here's what we're thinking. What do you think?'"
Why this matters for science and research
For researchers running preclinical studies, the immediate opportunity is in workflow automation-report generation, resource scheduling-where AI tools can cut weeks off timelines without touching the underlying science. The scientific capability, predicting a molecule's full toxicity profile from structure alone, remains unreliable. Data-sharing inertia among competitors keeps the predictive models starved of the negative results they need. Regulators are watching, and their acceptance will hinge on whether organizations can demonstrate that expert review remains the gatekeeper. The next move belongs to the consortia and contract research groups willing to broker the sharing of failure data. If they succeed, the toolset of preclinical development could finally expand beyond what any single company's data can show.
Your membership also unlocks:
