Fig. 1: Schematic overview of the data analysis pipeline.

Data from children with AS, NT children, and children with Dup15q were analyzed during wakefulness and NREM sleep. Children with AS generally have either genetic mutations or partial 15q deletions affecting the gene UBE3A, and an abnormal, high voltage delta EEG phenotype during both wakefulness (a_i) and NREM sleep (a_ii). NT children generally display low voltage, fast EEG activity during wakefulness (b_i) and high amplitude, slow activity during NREM sleep (b_ii). Children with Dup15q have partial trisomy or tetrasomy of 15q and an abnormal EEG phenotype characterized by fast β activity during wakefulness (c_i) and, to some extent, NREM sleep (c_ii). We extracted spectral (d) and entropy (e) features from wake and NREM sleep EEG recordings from the above cohorts. We then computed the mean across channels (a global average for single channel measures) or channel-pairs (a short-range or long-range average for functional connectivity measures) and subsequently z-scored these values (f). Next, we used two approaches for machine learning: feature selection was performed using linear mixed models (LMMs) to find features in each category that best differentiated wake from NREM sleep in AS as judged by regression coefficients (g). As an alternative, we also performed feature selection using PCA on the wakefulness—sleep feature contrast in AS and utilized the loadings from the first N PCs needed to explain ≥90% of the variance (h). In both approaches, we fit the regularized hyperparameter using 10-fold cross validation on AS data (i). Having determined this parameter, we then trained a regularized logistic regression classifier on AS data (j) and utilized two separate validations sets comprised of Dup15q (k) and NT (l) data. Finally, we repeated analysis steps h–l with the roles of NT and AS data switched (m), i.e., we trained classifiers on NT data and used Dup15q and AS data as validation sets. For channel-averaged EEG feature values and demographic variables, see Supplementary Data 1.