Abstract
QUANTITATIVE BEHAVIOURAL DYNAMICS OVER TIME IN MICE UNDER NATURALISTIC CONDITIONS
Madhu Nagathihalli Kantharajuand 2 co-authors
Max Delbrueck Center
FENS Forum 2026 (2026)
Barcelona, Spain
Presenter and authors
Presenter
Madhu Nagathihalli Kantharaju
Max Delbrueck Center
Co-authors
Rosalba Olga Proce; Hanna Hoernberg
Abstract
Social behaviour is fundamental to animal communication, cooperation, and group structure. Many social interactions are dyadic, occurring between pairs of individuals and forming the building blocks of more complex social systems.
Behavioural phenotyping in animal models is a key approach for studying neurological and physiological processes, and rigorous behavioural analysis is essential for understanding how the brain encodes social information. However, dyadic behavioural data are still commonly analysed manually or using supervised methods that depend heavily on researcher input. These approaches are time-intensive and prone to human bias, contributing to limited reproducibility in neuroscience. To address these limitations, we aim to develop a fully automated, accurate, and unbiased framework for quantifying dyadic behaviour.
We use DeepLabCut to track pairs of freely moving, age- and sex-matched mice and apply a rule-based clustering algorithm to identify interactions on a frame-by-frame basis. While this method enables classification of individual interaction frames, it does not capture the temporal organization of social behaviour, including behavioural sequences, transitions, and dynamics over time. This restricts our ability to study how social interactions unfold, escalate, or change within an episode. To overcome this limitation, we propose an algorithm that incorporates temporal analysis to characterise the structure and progression of dyadic social interactions, enabling a more comprehensive investigation of social behaviour.
Behavioural phenotyping in animal models is a key approach for studying neurological and physiological processes, and rigorous behavioural analysis is essential for understanding how the brain encodes social information. However, dyadic behavioural data are still commonly analysed manually or using supervised methods that depend heavily on researcher input. These approaches are time-intensive and prone to human bias, contributing to limited reproducibility in neuroscience. To address these limitations, we aim to develop a fully automated, accurate, and unbiased framework for quantifying dyadic behaviour.
We use DeepLabCut to track pairs of freely moving, age- and sex-matched mice and apply a rule-based clustering algorithm to identify interactions on a frame-by-frame basis. While this method enables classification of individual interaction frames, it does not capture the temporal organization of social behaviour, including behavioural sequences, transitions, and dynamics over time. This restricts our ability to study how social interactions unfold, escalate, or change within an episode. To overcome this limitation, we propose an algorithm that incorporates temporal analysis to characterise the structure and progression of dyadic social interactions, enabling a more comprehensive investigation of social behaviour.