B-SOiD

B-SOiD identifies and classifies spontaneous behaviors from DeepLabCut limb-localization pose-estimation data using an unsupervised machine-learning algorithm to enable quantitative analysis of movement kinematics.

Key Features:

• Unsupervised Learning: Discovers behavioral patterns from raw pose data without requiring labeled training datasets.
• Pose Pattern Analysis: Extracts spatiotemporal pose patterns from limb localization data and clusters them using novel statistical methods.
• Generalization Across Subjects and Labs: Generalizes learned behavioral categories across different subjects and experimental setups.
• Frameshift Alignment Paradigm: Implements a frameshift alignment paradigm to improve temporal resolution and facilitate alignment with electrophysiological recordings.
• Single Camera Utilization: Operates with single-camera pose-estimation inputs.
• Behavioral and Kinematic Measures: Outputs discrete sub-action categories and kinematic measures of individual limb trajectories.
• MATLAB Implementation: Algorithm implemented in MATLAB for execution and analysis.
• DeepLabCut Integration: Accepts pose-estimation outputs from DeepLabCut (limb localization) as input.

Scientific Applications:

• Naturalistic behavior quantification: Quantitative analysis of spontaneous and naturalistic animal behaviors.
• Neural–behavioral correlation: Temporal alignment of behavioral categories with electrophysiological recordings to correlate neural activity with actions.
• Movement disorder research: Analysis of movement-related pathologies including pain, obsessive-compulsive disorder (OCD), and other movement disorders.

Methodology:

Implemented in MATLAB, B-SOiD uses DeepLabCut limb-localization pose data to extract spatiotemporal pose patterns, clusters those patterns with novel statistical methods, and applies a frameshift alignment paradigm for temporal alignment with electrophysiological recordings.