Capturing silent oxidative stress in early Alzheimer's disease: prediction of CSF biomarkers from sleep qEEG data

Oxidative stress is a central pathogenic process in the earliest stages of Alzheimer's disease (AD), promoting non-enzymatic protein modifications that accumulate in cerebrospinal fluid (CSF) before measurable neurodegeneration. These alterations impair proteostasis and disrupt sleep-regulating neural circuits, producing characteristic changes in sleep electroencephalographic patterns. Because CSF sampling is invasive, quantitative electroencephalography (qEEG) has emerged as a promising non-invasive proxy for early oxidative processes. Here, we investigated whether nonlinear and time-domain sleep qEEG features can estimate CSF oxidative stress biomarkers in early AD using machine learning (ML) models. Forty-two mild-to-moderate AD patients underwent overnight polysomnography, from which sleep qEEG features were extracted. CSF protein oxidation biomarkers-glutamic semialdehyde, aminoadipic semialdehyde, N-carboxyethyl-lysine, N-carboxymethyl-lysine, and N-malondialdehyde-lysine-were quantified by gas chromatography/mass spectrometry, and ML models were trained to predict CSF biomarker levels from qEEG features. The best-performing model was a random forest trained on the first principal component, achieving an $R^2$ of 0.625 and a mean absolute error (MAE) of 467.1 pg/mL. Features derived from frontal and central electrodes during slow-wave sleep and rapid eye movement sleep contributed most strongly to predictive performance. Predictions for healthy controls displayed distributions distinct from those of AD patients, supporting the biological specificity of the qEEG-based estimates. These exploratory analyses suggest that sleep qEEG combined with ML can noninvasively capture silent oxidative processes involved early in AD pathological cascades, with potential for risk stratification, disease monitoring, and non-invasive upstream biomarkers development.