A novel machine learning system for identifying sleep-wake states in mice

Area of Science:

• Neuroscience
• Computational Biology
• Machine Learning

Background:

• Manual scoring of electroencephalogram (EEG) and electromyogram (EMG) recordings is time-consuming and subject to inter-rater variability.
• A four-state sleep-wake classification system (active wake, quiet wake, NREM, REM) offers greater precision for behavioral analysis but is complex to score manually.
• Machine learning holds potential for automating sleep-wake state classification due to characteristic physiological features.

Purpose of the Study:

• To develop and validate an automated machine learning system, SleepEns, for accurate sleep-wake behavior classification.
• To compare the performance of SleepEns against human expert scoring.
• To assess the impact of automated classification on sleep research efficiency.

Main Methods:

• Development of SleepEns, a novel time-series ensemble machine learning architecture.
• Training and validation of SleepEns using EEG and EMG recordings.
• Performance evaluation by comparing SleepEns classifications against expert human scorers.

Main Results:

• SleepEns achieved 90% accuracy compared to a primary expert, statistically similar to other human experts.
• Blind evaluation by the primary expert showed SleepEns had 99% acceptable performance, accounting for physiological classification variability.
• SleepEns classifications preserved essential sleep-wake characteristics, mirroring expert analyses.

Conclusions:

• SleepEns provides an accurate and efficient automated method for classifying sleep-wake states.
• The machine learning ensemble significantly reduces the time required for sleep-wake behavior analysis.
• SleepEns has the potential to advance sleep research in mice and humans by enabling faster and more consistent data analysis.