Comparative analysis of BORIS, Ethovision, DeepLabCut, and SimBA for quantifying autism spectrum disorder-like behaviors in the valproic acid mouse model.

Abstract

Preclinical research often relies on animal observation and subsequent behavioral analysis to study brain function; however, traditional methods are considered time-consuming and prone to human error. In contrast, emerging machine learning (ML) approaches now enable rapid, objective, and high-resolution behavioral assessment, such as DeepLabCut (DLC), combined with post-processing tools like Simple Behavioral Analysis (SimBA), which allow high-resolution behavioral classification. DLC provides accurate markerless tracking, while SimBA improves sensitivity and reliability in behavior identification. This study tests the hypothesis that pose-estimation-based behavioral analysis increases sensitivity for detecting functionally relevant impairments in social investigation and motor pattern organization in the valproic acid (VPA) mouse model of ASD, compared with conventional semi-automated tracking (Ethovision) and manual scoring (BORIS). Our results revealed significant and consistent core ASD-like symptoms in VPA-exposed mice across all methods. In the 3-chamber test, the tracking of the animal's nose provided greater precision and accuracy in detecting sociability deficits in VPA-exposed mice compared to the Ethovision analysis method. Correlation and Bland-Altman analyses indicated moderate agreement between both approaches for chamber time, but low concordance for the time in the proximity of the cages. Additionally, VPA-exposed mice exhibited significantly more repetitive behaviors (self-grooming and rearing) across both scoring methods. Indeed, DLC and BORIS scoring results demonstrated a higher correlation coefficient and a lower bias in the Bland-Altman analysis. Overall, this study demonstrates that integrating DLC and SimBA enhances behavioral scoring precision, overcomes limitations of conventional methods, and surpasses commercial automated tracking systems in detecting ASD-like phenotypes in mice.