VECTR-Clasp: An open machine-learning and vector-based framework for objective quantification of motor dysfunction during hind-limb clasping in Cdkl5-deficient mice

Hind-limb clasping is a widely used motor assay in rodent models of neurological disease, yet its scoring remains dependent on categorical, observer-defined scales that lack sensitivity to subtle kinematic features. Here, we present an integrated pipeline combining DeepLabCut for markerless pose estimation, SimBA for automated clasping detection, and VECTR-Clasp, an open-source R package, for continuous vector-based geometric analysis of movement during tail suspension. A SimBA random forest classifier trained on DeepLabCut pose tracks achieved automated clasping detection approaching human-level performance, with output closely matching the scoring intersection of two independent raters. Beyond binary classification, VECTR-Clasp extracted continuous circular and geometric measures, including head directionality, movement amplitude, and lateral swing frequency, from the same pose estimation data, revealing previously uncharacterised microphenotypes in Cdkl5-deficient mice. Knockout animals displayed reduced snout displacement, higher directional consistency, and fewer lateral swings compared to wildtype littermates, indicative of constrained or stereotyped movement patterns present even in the absence of overt clasping. These kinematic features were undetectable using traditional categorical scoring. VECTR-Clasp is fully open-source, compatible with standard DeepLabCut outputs, and generalisable to related suspended-mouse paradigms including the tail suspension test, providing a broadly applicable framework for continuous motor phenotyping across preclinical models. MotivationQuantitative assessment of motor behaviour in rodents remains constrained by categorical scoring systems that limit sensitivity, reproducibility, and the ability to detect subtle phenotypes. We developed VECTR-Clasp to address these limitations by introducing a vector-based geometric framework that transforms standard pose estimation outputs into continuous, body-relative kinematic representations. By combining DeepLabCut for pose estimation, SimBA for automated clasping classification, and VECTR-Clasp for downstream geometric analysis, our pipeline moves beyond binary event detection to extract movement features invisible to traditional scoring. Applied to Cdkl5-deficient mice, this integrated approach reveals previously uncharacterised motor microphenotypes, demonstrating that computational behavioural analysis can uncover biologically meaningful phenotypic structure beyond what categorical scales can resolve. HighlightsO_LIDeepLabCut-SimBA pipeline automates hind-limb clasping detection at human-level accuracy C_LIO_LIVECTR-Clasp extracts continuous geometric and circular kinematics from pose estimation data C_LIO_LICdkl5-deficient mice show constrained snout trajectories and reduced lateral swinging during suspension. C_LIO_LIKinematic microphenotypes are detectable in knockout mice even in the absence of overt clasping C_LI