IntelliSleepScorer, a software package with a graphic user interface for automated sleep stage scoring in mice based
Lei A Wang1, Ryan Kern2, Eunah Yu1
1Stanley Center for Psychiatric Research, Broad Institute of MIT and Harvard, 75 Ames Street, Cambridge, MA, 02142, USA.
Scientific Reports
|March 16, 2023
Summary
IntelliSleepScorer offers an automated pipeline for mouse sleep stage scoring using LightGBM machine learning. This validated tool achieves high accuracy, comparable to human experts, providing a reliable solution for sleep research.
Area of Science:
- Neuroscience
- Computational Biology
- Bioengineering
Background:
- Automated sleep scoring is crucial for rodent research but lacks validated public pipelines.
- Machine learning, particularly electroencephalogram (EEG) and electromyogram (EMG) analysis, shows promise for sleep stage classification.
- Current methods often require manual scoring or lack generalizability across different experimental conditions.
Purpose of the Study:
- To develop and validate a user-friendly, automated software package for accurate sleep stage scoring in mice.
- To implement a machine learning model that achieves high performance and generalizability for analyzing sleep EEG and EMG data.
- To provide a publicly available, out-of-the-box solution for researchers studying sleep in rodent models.
Main Methods:
- Developed IntelliSleepScorer, a software package utilizing the Light Gradient Boosting Machine (LightGBM) algorithm for automatic sleep stage classification (NREM, REM, wake).
- Trained and validated LightGBM models on a large dataset comprising 5776 hours of mouse sleep EEG and EMG signals from 519 recordings.
- Evaluated model performance using accuracy and Cohen's kappa, comparing it against baseline models and human expert scoring, and tested generalizability on independent datasets.
Main Results:
- The LightGBM model achieved a high overall accuracy of 95.2% and a Cohen's kappa of 0.91, outperforming logistic regression and random forest models.
- Model performance was comparable to human expert scoring across different sleep stages and demonstrated robustness across varied datasets, sampling frequencies, and light/dark cycles.
- Validated generalizability on independent datasets (kappa >= 0.80), lower sampling frequencies (kappa = 0.90), and with reduced electrode configurations (kappa >= 0.89).
Conclusions:
- The LightGBM models integrated into IntelliSleepScorer provide state-of-the-art performance for automatic sleep stage scoring in mice.
- IntelliSleepScorer offers a validated, generalizable, and user-friendly solution, addressing the need for reliable automated sleep analysis in rodent research.
- This publicly available software facilitates efficient and accurate sleep phenotyping, advancing the field of sleep science.
