B-SOiD, an open-source unsupervised algorithm for identification and fast prediction of behaviors.

Area of Science:

• Animal behavior analysis
• Machine learning in biology
• Biomechanical analysis

Background:

• Studying naturalistic animal behavior is challenging.
• Limb localization via machine learning is advancing, but behavior extraction requires spatiotemporal analysis.
• Existing methods lack speed, generalizability, and unbiased behavioral identification.

Purpose of the Study:

• To develop an unsupervised algorithm linking animal poses to actions and kinematics.
• To create an open-source tool for unbiased, efficient, and generalizable behavior identification.
• To overcome temporal resolution limitations in behavioral analysis.

Main Methods:

• Developed B-SOiD, an unsupervised algorithm using machine classification on clustered pose pattern statistics.
• Employed a frameshift alignment paradigm for enhanced temporal resolution.
• Utilized a single, off-the-shelf camera for data acquisition.

Main Results:

• Achieved significantly improved processing speed and generalizability across subjects and labs.
• Enabled identification of sub-action categories and kinematic measures for individual limb trajectories.
• Provided a method to link poses to actions and kinematics without user bias.

Conclusions:

• B-SOiD offers a powerful, open-source solution for detailed animal behavior and kinematic analysis.
• The algorithm is critical for studying models of pain, OCD, and movement disorders in various animal models.
• B-SOiD overcomes previous limitations in speed, generalization, and temporal resolution for behavioral studies.