Machine Learning Classification of Purkinje Cell Spike Features for Electrophysiological Characterization and Potential ASD Relevance

Abstract

Purkinje cells, as principal output neurons of the cerebellar cortex, critically influence motor coordination and cognitive functions. Abnormal firing patterns in these cells have been implicated in autism spectrum disorder (ASD), highlighting their potential as biomarkers for neurological dysfunction. In this study, we developed a machine learning-based classification pipeline utilizing spike-derived electrophysiological features to differentiate between experimental conditions in mouse Purkinje cells. Whole-cell patch-clamp recordings from three distinct experimental groups—AA-only, PE-only and Control—were analyzed to extract firing rate, mean inter-spike interval (ISI), ISI variability (standard deviation), and coefficient of variation (CV_ISI). We compared the performance of several machine learning models, including support vector machines (SVM), k-nearest neighbors (KNN), logistic regression, and random forest (RF). Among them, RF achieved the highest accuracy (67%) in 3-fold cross-validation. Feature importance analysis showed that CV_ISI and firing rate were the most predictive features. These findings suggest that spike statistics carry valuable information for ASD-related neural classification and provide a foundation for interpretable, data-driven analysis of cerebellar electrophysiology.