Fig. 1

Overview of motor stratification and kinematic data analysis workflow. Panel A shows how IRCCS-MEDEA and NDA datasets are combined and how the initial preprocessing steps are implemented to remove confounding effects of originating study ID, MABC2 module, and sex. Panel B shows the workflow for stability-based relative clustering validation (reval) analyses that aimed to identify the optimal number of clusters (best k) that minimizes normalized stability in independent splits of the data (training and validation) and estimate the generalization accuracy of such optimal (best k) clustering solution. Panel C shows the analysis workflow for how we estimated motor noise from kinematics data acquired during a simple reach-to-drop task from the IRCCS-MEDEA dataset. Within this task, 10 repeat trials were administered and we used multivariate dynamic time warping (DTW) to align and compare motor kinematic trajectories across repeat trials. Motor noise is operationalized as the median similarity across trials (DTW dist norm) whereby higher estimates are indicative of more motor noise (i.e., increased dissimilarity between repeat trials). Panel D indicates the final step of hypothesis testing for subtype differences in motor noise