Somnotate: A probabilistic sleep stage classifier for studying vigilance state transitions

Area of Science:

• Neuroscience
• Computational Biology
• Sleep Research

Background:

• Electrophysiological recordings are crucial for sleep research, generating vast datasets requiring accurate sleep stage annotation (polysomnography).
• Current methods often overlook intermediate vigilance states, missing vital information on sleep-wake dynamics.
• These intermediate states are critical for understanding the nuances of sleep architecture.

Purpose of the Study:

• To develop an advanced computational method for accurate sleep stage annotation, including intermediate states.
• To leverage annotation certainty to analyze sleep transition dynamics.
• To provide an open-source tool for enhanced sleep research.

Main Methods:

• Developed Somnotate, a probabilistic classifier combining linear discriminant analysis (LDA) and a hidden Markov model (HMM).
• Validated Somnotate on mouse electrophysiological data, assessing accuracy, robustness, and compatibility.
• Utilized annotation certainty quantification to analyze intermediate vigilance states and transition patterns.

Main Results:

• Somnotate achieved annotation accuracies exceeding human experts on mouse electrophysiological data.
• The classifier demonstrated robustness to training data errors and compatibility with various recording setups.
• Analysis revealed intermediate states related to successful and failed sleep transitions, differentially affected by experimental manipulations.

Conclusions:

• Somnotate sets a new standard for polysomnography, offering superior accuracy and insights into sleep dynamics.
• The quantification of annotation certainty provides a novel approach to studying sleep stage transitions.
• This open-source tool has the potential to significantly advance sleep research by revealing mechanisms underlying sleep-wake dynamics.