South San Francisco, California, August 6, 2026
Olio Labs has presented a new AI-driven in vivo drug discovery platform designed to predict human clinical outcomes from just 24 hours of rodent behavioral data, potentially providing pharmaceutical researchers with a faster and more information-rich approach to preclinical development. The research addresses a longstanding challenge in drug development: conventional animal models frequently struggle to predict how investigational therapies will ultimately perform in humans. According to the study, Olio Labs developed a scalable home-cage system that continuously records animal behavior and generates thousands of quantitative features, which are then connected with human clinical trial outcomes through computational models. The platform was evaluated for its ability to predict outcomes including gastrointestinal adverse events, cardiac toxicity, neuropsychiatric side effects and long-term weight loss.
AI Connects Rodent Behavior With Human Clinical Outcomes
At the core of the approach is a purpose-built phenotype acquisition device, or PAD, that enables continuous and largely unattended behavioral monitoring. A fleet of 96 PADs can generate 2,304 hours of video every day, while computer-vision models analyze animal pose, movement, location, social activity and other behaviors to create high-dimensional behavioral signatures. The experimental workflow also incorporates QR-code-based animal and syringe tracking, automated injection-quality assessment and time-stamped records. According to the researchers, these features are intended to strengthen data integrity, experimental standardization and chain of custody, with the system engineered around standards underpinning Good Laboratory Practice. Rather than selecting a small number of predefined animal behaviors as proxies for human outcomes, the platform trains models directly against curated clinical trial information. Human trial records, published studies and other clinical data are structured to capture doses, endpoints and adverse events, while corresponding drug-induced mouse behavioral fingerprints are used to develop cross-species predictive models.
Platform Predicts Toxicity and Treatment Tolerability
The researchers initially investigated whether behavioral patterns could predict gastrointestinal adverse-event rates associated with therapies studied for obesity and diabetes. Models were developed for clinically relevant outcomes including nausea, vomiting, diarrhea and constipation, with the study reporting strong relationships between predicted and observed human outcomes. The platform was also tested against drugs associated with neuropsychiatric and cardiac safety concerns. Researchers reported that behavioral models identified clinical failures associated with neuropsychiatric effects and detected signals connected with cardiac toxicity, including hERG-associated liabilities. Importantly, the platform is not presented as simply interpreting whether an animal displays one particular symptom. Instead, it evaluates large numbers of behavioral measurements simultaneously, allowing models to identify combinations of features carrying clinically relevant information. This approach could help researchers evaluate efficacy and tolerability together, an important consideration when selecting drug candidates for further development.
Advertisement
24-Hour Predictions Could Accelerate Preclinical R&D
One of the study’s most notable findings involved prediction of human weight-loss outcomes. Traditional rodent weight-loss experiments commonly run for two to three weeks, but Olio Labs compared those conventional measurements with predictions generated from only 24 hours of behavioral information. The researchers reported that the single-day behavioral approach showed a stronger relationship with human clinical data and reduced median absolute prediction error by approximately 70% compared with conventional longitudinal weight tracking. The platform was also used in rapid iterative experiments in which candidate compounds were evaluated for predicted efficacy and gastrointestinal tolerability on a daily cycle, illustrating how in vivo experimentation could potentially operate more like high-throughput screening. The researchers emphasize that the technology is intended to complement rather than simply replace conventional confirmatory animal studies, particularly when development teams need to rank multiple candidates, doses or therapeutic profiles. However, the study also acknowledges important limitations: predictions depend on the quality and consistency of human clinical data used for training, and cross-species biological differences can limit performance when relevant drug targets or pathways are not sufficiently conserved in mice. Overall, the research suggests that combining AI, computer vision, continuous behavioral phenotyping and clinical trial data could provide pharmaceutical developers with a new strategy for improving translational decision-making and identifying promising drug candidates earlier in development.
Source: Olio Labs press release
