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

Abstract

Quantitative assessment of motor behaviour in rodents is central to the study of neurological disease, yet it remains constrained by manual categorical scoring systems that limit sensitivity, reproducibility, and the ability to detect subtle phenotypes. The hind-limb clasping assay, a standard test of motor dysfunction across many mouse models, is typically scored on an observer-defined categorical scale that may overlook meaningful movement features. We developed VECTR-Clasp, an open vector-based geometric framework that transforms standard pose-estimation output into continuous, body-relative kinematic measures. Combining DeepLabCut for markerless pose estimation with SimBA for automated clasping classification, our pipeline first reproduces conventional clasping detection at a level of agreement approaching that between trained human raters, and then extracts continuous geometric descriptors, including head directionality, total distance travelled, and swing count, that are inaccessible to categorical scoring. Applied to a mouse model of CDKL5 deficiency disorder, a rare neurodevelopmental condition in which motor impairment is a core clinical feature, this approach revealed previously uncharacterised motor microphenotypes: affected animals showed more constrained head direction, reduced overall movement, and fewer swings than wildtype controls. Critically, these differences were also present in affected animals that displayed no overt clasping. Together, these findings demonstrate that continuous geometric analysis of pose-estimation data uncovers motor phenotypes beyond the resolution of categorical scales, providing a more sensitive and reproducible framework for quantifying motor dysfunction. As it operates on standard pose-estimation output, the approach extends readily to other behavioural assays, disease models, and the evaluation of therapeutic interventions. Author SummaryMany neurological disorders impair movement, and researchers frequently study these deficits in mouse models. A widely used test lifts a mouse by the tail and records whether it retracts its hind limbs towards its body, a reflex associated with motor dysfunction. Traditionally, an observer watches the animal and assigns a categorical score. This approach depends on human judgement, varies between raters, and can miss subtle movement differences. We developed VECTR-Clasp, an open-source pipeline that validates machine-learning pose tracking to automate traditional clasping assessment and extends current methods using vector-based geometry. This quantifies the direction and extent of head and body movements, providing additional motor readouts. In a mouse model of CDKL5 deficiency disorder, a rare paediatric neurological disorder, VECTR-Clasp matched human scoring while revealing kinematic differences missed by standard assessment, including in animals appearing unaffected. This objective, sensitive framework could detect subtle motor changes and treatment responses across neurological diseases.