Development and applications of a machine learning model for an in-depth analysis of pentylenetetrazol-induced seizure-like behaviors in adult zebrafish

Abstract

Epilepsy, a neurological disorder causing recurring seizures, is often studied in zebrafish by exposing animals to pentylenetetrazol (PTZ), which induces clonic- and tonic-like behaviors. While adult zebrafish seizure-like behaviors are well characterized, manual assessment remains challenging due to its time-consuming nature, potential for human error/bias, and the risk of overlooking subtle behaviors. Aiming to circumvent these issues, we developed a machine learning model for automating the analysis of subtle abnormal and seizure-like behaviors in PTZ-exposed adult zebrafish. To improve pharmacological validity, we also evaluated the efficacy of two anticonvulsant drugs, diazepam (DZP) and valproate (VALP). As strategy, we employed a Random Forest algorithm combined with a post-processing analysis to identify six behavioral phenotypes in PTZ-exposed zebrafish. We found a concentration-dependent effect of PTZ and a distinct behavioral phenotype for DZP and VALP, where these drugs showed different protective profiles. Altogether, our novel data highlights the use of machine learning models to better understand complex behavioral phenotypes associated to PTZ-induced seizures. The ability to detect frame-by-frame and distinct actions of anticonvulsant drugs provides new perspectives on measuring seizure-like responses, as well as possible therapeutic strategies. The approach used here constitutes an important leap on behavioral analysis that can accelerate the discovery of new treatments for seizure disorders.

Keywords: Anticonvulsant drugs; Clonic-like behavior; Diazepam; Random Forest Model; Tonic-like behavior; Valproate.