Machine Learning-Enabled Digital Toxicology for Enhanced Discovery and Decisions

Machine learning-defined digital measures applied to home cage computer video detected and quantitatively characterized dose-responsive changes in activity in mice and rats given single doses of caffeine or chlorpromazine. The onset and duration of effect was observationally and statistically identifiable in individual animals as a deviation from time-matched baseline activity. Chemically-induced seizures induced by pentylenetetrazole (PTZ) were also reliably detected with a digital measure for LORR.

The objective and quantitative measures revealed by digital sensors and ML-informed algorithms of animal behavior can significantly contribute to the primary aims of a toxicology study by detecting a test article-related effect as well as informing its exposure relationship, character, magnitude, duration, reversibility, adversity, and monitorability. A more dynamic and temporal characterization also allows a better integration with the plasma toxicokinetics of the test article providing insights into potential modes of action and likelihood of accommodation or progression.