AUTOMATED BEHAVIORAL ANALYSIS IN PARKINSONIAN MODELS: FROM DYSKINESIA CLUSTERING TO DOPAMINERGIC MODULATIONS FINGERPRINTS

Dopamine plays a central role in initiating, sequencing, and suppressing actions. In Parkinson’s disease (PD), loss of dopaminergic neurons causes bradykinesia and rigidity, while dopamine replacement with L-DOPA often leads to abnormal involuntary movements known as L-DOPA-induced dyskinesia (LID). The heterogeneity of LID reflects underlying circuit dysfunctions that remain poorly understood, partly due to limitations in behavioral measurement.
To address this, we developed a semi-automated pipeline for high-resolution classification of dyskinesia in freely moving mice (Alcacer et al., Cell Reports, 2025). Using video, head-mounted inertial sensors (IMUs), and unsupervised clustering, we identified two dyskinesia subtypes and a novel pathological rotation phenotype. Calcium imaging revealed that each behavior was encoded by specific ensembles of hyperactive D1- and D2-spiny projection neurons, many also active during normal movements -suggesting that L-DOPA exaggerates existing circuit motifs.
Expanding beyond LID, we now ask how selective dopaminergic agents - targeting D1 vs. D2 receptors - reshape behavior across dose and disease state. Using open-field video (with mirrored views) and IMUs from intact and 6-OHDA-lesioned mice treated with L-DOPA, SKF-38393, or quinpirole, we aim to automatically extract drug- and dose-specific behavioral fingerprints with sub-second resolution.
Our analysis combines supervised and unsupervised machine learning (SVM, XGBoost, A-SOiD) to decode treatment and identify fine-grained motor signatures. This work bridges pharmacology, behavior, and neural dynamics, establishing standardized metrics aligned with human wearable data—and may ultimately support future AI-assisted treatment strategies in PD, paralleling advances in closed-loop insulin delivery for diabetes.