Related Experiment Video
Updated: Aug 18, 2025
08:51
Measuring Respiratory Function in Mice Using Unrestrained Whole-body Plethysmography
Published on: August 12, 2014
30.7K
Machine learning-based clustering and classification of mouse behaviors via respiratory patterns
Emma Janke1, Marina Zhang2, Sang Eun Ryu3
1Department of Neuroscience, University of Pennsylvania Perelman School of Medicine, Philadelphia, PA 19104, USA.
Iscience
|December 8, 2022
Summary
Breathing patterns vary significantly across different mouse behaviors, from spontaneous actions to stress and fear responses. These distinct respiratory dynamics can be identified using machine learning, revealing a strong link between breathing and behavior.
Area of Science:
- Neuroscience
- Behavioral Biology
- Respiratory Physiology
Background:
- Breathing is influenced by metabolic and emotional factors.
- Rodents are key models for studying respiratory control.
- Systematic characterization of breathing across rodent behaviors is lacking.
Purpose of the Study:
- To systematically characterize breathing dynamics across diverse mouse behaviors.
- To investigate the relationship between respiratory patterns and behavioral states.
- To assess the utility of breathing patterns for behavior classification.
Main Methods:
- Direct intranasal pressure recordings were used for detailed respiratory data.
- K-means clustering grouped behavioral states based on respiratory features.
- RUSBoost (random undersampling boost) classification was employed for supervised learning.
Main Results:
- A wide diversity of breathing patterns was observed across spontaneous, odor-, stress-, and fear-induced behaviors.
- Eleven behavioral states were clustered into four groups with distinct respiratory signatures.
- Breathing patterns accurately classified behaviors with 80% accuracy using machine learning.
Conclusions:
- Breathing patterns are tightly linked to behavioral states in mice.
- Respiratory dynamics can aid in distinguishing between similar behaviors.
- Breathing patterns offer insights into the internal states associated with behaviors.
