Deep learning based neurobehavior assessment in rodent model of isolation stress

Background and objectives: Social isolation is a potent psychosocial stressor that alters locomotor, exploratory, anxiety-like, and cognitive behaviours. Conventional manual scoring is subjective and often fails to detect subtle behavioral changes. This study aimed to implement an automated deep learning–based behavioral phenotyping pipeline using DeepLabCut and SimBA to objectively quantify isolation-induced behavioral deficits and to assess the modulatory effects of drug treatment with high precision and reproducibility. Key findings: Isolated rats showed significant reductions in locomotion, rearing, center exploration, and novel object interaction, indicating impaired exploratory and cognitive functions. Unsupervised clustering and heatmap analyses revealed distinct behavioral phenotypes across control, isolated, and drug-treated groups. The deep learning workflow enabled accurate pose estimation, behaviour annotation, and multivariate profiling. Drug treatment partially reversed isolation-induced deficits, demonstrating the sensitivity of this framework for detecting therapeutic effects. Conclusion and future perspectives: Social isolation induces robust neurobehavioral impairments that can be reliably captured using automated deep learning–based analysis. This approach offers a scalable, unbiased alternative to manual scoring. Future integration of CNN-based astrocyte segmentation will enable multiscale correlation of cellular and behavioral changes, advancing mechanistic understanding of stress-induced neuropathology and treatment responses.