Abstract
Manual scoring of longitudinal electroencephalographical (EEG) data is a slow and time-consuming process. Current advances in the application of machine-learning and artificial intelligence to EEG data are moving fast; however, there is still a need for expert raters to validate scoring of EEG data. We hypothesized that power-frequency profiles are determining the state and 'set the framework' for communication between neurons. Based on these assumptions, a scoring method with a set frequency profile for each vigilance state, both in sleep and awake, was developed and validated. We defined seven states of the functional brain with unique profiles in terms of frequency-power spectra, coherence, phase-amplitude coupling, α exponent, functional excitation-inhibition balance (fE/I), and aperiodic exponent. The new method requires a manual check of wake-sleep transitions and is therefore considered semi-automatic. This semi-automatic approach showed similar α exponent and fE/I when compared to traces scored manually. The new method was faster than manual scoring, and the advanced outcomes of each state were stable across datasets and epoch length. When applying the new method to the neurexin-1α (Nrxn1α) gene deficient mouse, a model of synaptic dysfunction relevant to autism spectrum disorders, several genotype differences in the 24-h distribution of vigilance states were detected. Most prominent was the decrease in slow-wave sleep when comparing wild-type mice to Nrxn1α-deficient mice. This new methodology puts forward an optimized and validated EEG analysis pipeline for the identification of translational electrophysiological biomarkers for brain disorders that are related to sleep architecture and E/I balance.