Tracking changes in behavioural dynamics using prediction error

Area of Science:

• Ethology
• Computational Biology
• Neuroscience

Background:

• Automated video analysis generates complex time-series data of organismal movement.
• Characterizing behavioral dynamics requires advanced quantitative tools.
• The nematode Caenorhabditis elegans offers a well-established model for pose and motion analysis.

Purpose of the Study:

• To develop a data-driven method for quantifying subtle behavioral variations at high temporal resolution.
• To reveal dynamic changes in behavior using prediction error relative to a reference behavioral attractor.
• To tailor behavioral analysis to specific research questions at individual or group levels.

Main Methods:

• Empirical dynamic modeling to construct a time-delay-embedded attractor from reference behavioral data.
• Quantification of behavioral change as prediction error with respect to the attractor.
• Validation using movement dynamics of Caenorhabditis elegans during specific maneuvers and responses.

Main Results:

• Successfully detected subtle changes in C. elegans movement dynamics during delta turns.
• Tracked the return to baseline behavior post-aversive stimulus in individual worms.
• Revealed variations in escape response dynamics among individual worms.

Conclusions:

• The proposed method effectively quantifies dynamic behavioral changes using reference attractors and prediction error.
• This approach is broadly applicable to behavioral researchers analyzing video-derived time-series data.
• Offers a powerful tool for dissecting complex behaviors in various organisms.