Sleep Identification Enabled by Supervised Training Algorithms (SIESTA): An Open-Source Platform for Automatic Sleep...

Area of Science:

• Neuroscience
• Computational Biology
• Sleep Science

Background:

• Accurate polysomnographic sleep stage scoring is vital for studying sleep function and disorders.
• Manual scoring of electrocorticography (ECoG) and electromyography (EMG) recordings is laborious and time-consuming, especially for long-duration studies.
• Current methods are impractical for large experimental cohorts and circadian rhythm research.

Purpose of the Study:

• To develop an automated, open-source tool for accurate sleep stage detection in rodents.
• To provide a user-friendly solution for analyzing large datasets of ECoG and EMG recordings.
• To facilitate sleep research in rodent models by overcoming manual scoring bottlenecks.

Main Methods:

• Developed SIESTA, an open-source Python toolkit utilizing a supervised machine learning algorithm.
• SIESTA employs a hierarchical classifier based on logistic regression to score sleep stages.
• Features are extracted from ECoG and EMG signals for automated detection of wakefulness, REM, and NREM sleep.

Main Results:

• SIESTA achieved high mean F1 scores: 0.94 for wakefulness, 0.94 for NREM sleep, and 0.74 for REM sleep.
• The toolkit was validated on data from wild-type mice, mutant mouse lines, and rats under various conditions.
• External validation with manually scored data from three independent laboratories confirmed SIESTA's accuracy.

Conclusions:

• SIESTA offers an accurate and efficient automated solution for sleep stage scoring in rodents.
• The open-source nature and user-friendly interface make SIESTA accessible to researchers without coding expertise.
• This toolkit is expected to significantly advance sleep research in rodent models by enabling large-scale data analysis.