Development and application of machine learning-based digital biomarkers measure to detect loss of upright posture associated with spontaneous seizures in preclinical epilepsy models

Epilepsy is a chronic neurological disorder characterized by seizures and periods of unusual behaviors that affects an estimated 50 million people worldwide. Rodent models of epilepsy are essential for understanding underlying mechanisms of the disorder and developing novel therapeutics. The gold standard assay to monitor for spontaneous seizures in rodent models of epilepsy is video/ electroencephalography (vEEG). vEEG in animals is low throughput, requiring specialized data acquisition systems, surgical implantation of electrodes, and expert data analysis. For these reasons, the field is often limited in its ability to fully characterize spontaneous seizure dynamics across a growing number of preclinical epilepsy models.

This work demonstrates advances in the development of automated detection of seizures of individual animals in group-housed mice. Continuous, objective, and quantitative assessment of defined behaviors was done using machine learning-enabled algorithms applied in real time to raw computer vision video. For development of a lateral position measure, mice were treated with pentylenetetrazol (PTZ). The model was further developed using a natural history characterization of two mouse models of Dravet Syndrome, a severe genetic epileptic encephalopathy, characterized by spontaneous tonic-clonic seizures and SUDEP over the early course disease. Furthermore, we demonstrate how digital biomarkers allow for multiplexing of phenotypic readouts, assessing the interplay between seizure, activity metrics, sleep wake cycle, etc. concurrently in the same cohort of mice. The ability to noninvasively detect lateral position due to loss of upright posture n the home cage in real time offers value for health and welfare monitoring. The non-invasive nature of digital home cage monitoring also enhances the efficiency of spontaneous seizure detection and offers significant value in development models that are more translationally relevant and reproducible. This data emphasizes the potential wide-ranging impacts of this technology in preclinical epilepsy research and beyond.