Machine Learning EEG Biomarkers in SYNGAP1 Rodent Models

SYNGAP1-related disorder (SRD) is a neurodevelopmental disorder caused by mutations in the SYNGAP1 gene. SRD patients experience seizures and developmental problems. Novel treatments for SRD are in development, but reliable biomarkers for tracking their clinical efficacy have yet to be identified. We construct a machine learning EEG classification pipeline in a SYNGAP1 rat model (9 SYNGAP1, 9 wild-type). Over 1000 spectral, complexity, and functional connectivity features were extracted and evaluated using XGBoost and CatBoost across four independently trained brain-state conditions: all epochs, rapid-eye-movement (REM), non-REM (NREM), and wake. Aggregating all epoch-level features to the animal-level with XGBoost achieved global area-under-ROC-curve of 0.963 and accuracy of 0.778 (p $<$ 0.001, 1000-permutation test). Top discriminating features were an enhanced delta-band cross-correlation between motor and visual cortex and reduced alpha/delta power ratio in motor cortex in SYNGAP1 animals, consistent with known EEG signatures of the disorder.