This study explores the use of machine learning (ML) to derive brain-specific biomarkers from electroencephalography (EEG) signals. The authors present a framework that conceptualizes the impact of peripheral signals and artifact removal techniques on ML models for age and sex prediction. Using over 2600 EEG recordings from public databases, they found that basic artifact rejection improved model performance, while further removal of peripheral signals using independent component analysis (ICA) decreased performance. The study reveals that peripheral signals can enable age and sex prediction but only explain a fraction of the performance provided by brain signals. The results suggest that careful handling of peripheral signals is necessary to develop CNS-specific biomarkers using ML. The authors also highlight the importance of considering the predictive value of peripheral signals, which can contain information about the outcome. Their findings provide insights into the practical and theoretical aspects of ML in EEG biomarker discovery, emphasizing the need for explicit artifact removal to ensure the interpretability and reliability of brain-specific biomarker models.This study explores the use of machine learning (ML) to derive brain-specific biomarkers from electroencephalography (EEG) signals. The authors present a framework that conceptualizes the impact of peripheral signals and artifact removal techniques on ML models for age and sex prediction. Using over 2600 EEG recordings from public databases, they found that basic artifact rejection improved model performance, while further removal of peripheral signals using independent component analysis (ICA) decreased performance. The study reveals that peripheral signals can enable age and sex prediction but only explain a fraction of the performance provided by brain signals. The results suggest that careful handling of peripheral signals is necessary to develop CNS-specific biomarkers using ML. The authors also highlight the importance of considering the predictive value of peripheral signals, which can contain information about the outcome. Their findings provide insights into the practical and theoretical aspects of ML in EEG biomarker discovery, emphasizing the need for explicit artifact removal to ensure the interpretability and reliability of brain-specific biomarker models.