Hierarchical Compression Reveals Sub-Second to Day-Long Structure in Larval Zebrafish Behavior

Area of Science:

• Animal Behavior
• Computational Neuroscience
• Ethology

Background:

• Animal behavior exhibits complex dynamics across various timescales.
• Understanding these multi-timescale dynamics requires advanced analytical tools.
• Larval zebrafish serve as a model organism for studying fundamental behavioral principles.

Purpose of the Study:

• To develop computational tools for capturing multi-timescale structure in behavioral data.
• To analyze the behavior of hundreds of larval zebrafish across multiple 24-hour cycles.
• To reveal the organization of behavior at timescales ranging from milliseconds to days.

Main Methods:

• Continuous tracking of hundreds of larval zebrafish over multiple day/night cycles.
• Extraction and analysis of millions of movements and pauses (bouts).
• Application of unsupervised learning for behavioral module and motif identification.
• Hierarchical compression to identify recurrent behavioral patterns.

Main Results:

• Behavior was reduced to sequences of active and inactive bout types (modules).
• Recurrent behavioral patterns (motifs) were identified through hierarchical compression.
• Module and motif usage demonstrated significant variation across the day/night cycle.
• Analysis successfully uncovered novel phenotypes in pharmacological and genetic mutants.

Conclusions:

• Larval zebrafish behavior is organized across multiple timescales, from sub-second to day-long.
• The developed computational tools provide a robust method for analyzing large-scale behavioral datasets.
• This approach offers new avenues for identifying behavioral phenotypes in response to genetic or environmental factors.