Keypoint MoSeq

Keypoint MoSeq parses continuous keypoint tracking data into discrete behavioral modules ('syllables') using an unsupervised machine learning approach that distinguishes noise from behavior.

Key Features:

• Noise-Behavior Discrimination: Uses a generative model to distinguish keypoint noise from behavior, enabling identification of syllables whose boundaries correspond to natural sub-second discontinuities in pose dynamics.
• Cross-Species Generalization: Works across multiple species including mice and fruit flies.
• Multi-Timescale Analysis: Generalizes beyond the syllable timescale to identify fast sniff-aligned movements and a spectrum of oscillatory behaviors.
• Superior Transition Detection: Outperforms commonly used alternative clustering methods at identifying transitions between actions.
• Neural-Behavioral Correlation: Captures correlations between neural activity and behavior.
• Behavior Classification: Classifies solitary or social behaviors in accordance with human annotations.

Scientific Applications:

• Behavioral Module Identification: Identifies behavioral syllables from keypoint data without human supervision.
• Neuroscience Research: Captures correlations between neural activity and behavior for studying neural-behavioral relationships.
• Social Behavior Analysis: Classifies social behaviors in accordance with human annotations.
• Species-Comparative Behavior: Applies across multiple species and timescales for comparative behavioral studies.
• Fast Movement Analysis: Identifies fast sniff-aligned movements in mice.
• Oscillatory Behavior Characterization: Identifies a spectrum of oscillatory behaviors in fruit flies.

Methodology:

Keypoint MoSeq uses a generative model to distinguish keypoint noise from behavior and identify syllables whose boundaries correspond to natural sub-second discontinuities in pose dynamics.