A lexical approach for identifying behavioral action sequences

Abstract

Animals display characteristic behavioral patterns when performing a task, such as the spiraling of a soaring bird or the surge-and-cast of a male moth searching for a female. Identifying such conserved patterns occurring rarely in noisy behavioral data is key to understanding the behavioral response to a distributed stimulus in unrestrained animals. Existing models seek to describe the dynamics of behavior or segment individual locomotor episodes rather than to identify occasional, transient irregularities that make up the behavioral response. To fill this gap, we develop a lexical, hierarchical model of behavior. We designed an unsupervised algorithm called “BASS” to efficiently identify and segment conserved behavioral action sequences transiently occurring in long behavioral recordings. When applied to navigating larval zebrafish, BASS extracts a dictionary of remarkably long, non-Markovian sequences consisting of repeats and mixtures of slow forward and turn bouts. Applied to a novel chemotaxis assay, BASS uncovers conserved chemotactic strategies deployed by zebrafish to avoid aversive cues consisting of sequences of fast large-angle turns and burst swims. In a simulated dataset of soaring gliders climbing thermals, BASS finds the spiralling patterns characteristic of soaring behavior. In both cases, BASS succeeds in identifying action sequences that are highly conserved but transient in the behavior deployed by freely moving animals. BASS can be easily incorporated into the pipelines of existing behavioral analyses across diverse species, and even more broadly used as a generic algorithm for pattern recognition in low-dimensional sequential data.

Maximally predictive ensemble dynamics from data

“…Similar ideas have been used previously [83], where the notion of relatively coherent sets has been introduced. We note that our metastate subdivisioning process proceeds from the longest-lived states down rather than from the shortest-lived states up, where the latter is more common in temporal clustering approaches [84–86].…”

High-throughput automated methods for classical and operant conditioning of Drosophila larvae

“…Temporally precise action reinforcement requires real-time identification of larval behaviours. Numerous algorithms already exist for real-time tracking of freely moving animals ( Stowers et al, 2017 ; Krynitsky et al, 2020 ; Mischiati et al, 2015 ; Fry et al, 2008 ; Straw et al, 2011 ; Swierczek et al, 2011 ) and subsequent offline behaviour analysis ( Mathis et al, 2018 ; Veeraraghavan et al, 2008 ; Dankert et al, 2009 ; Robie et al, 2017 ; Mirat et al, 2013 ; Reddy et al, 2020 ; Stephens et al, 2008 ; Gupta and Gomez-Marin, 2019 ). The Drosophila larva has been of notable analytic interest due its deformable body and limited set of distinguishing physical features.…”

Maximally predictive ensemble dynamics from data

“…Similar ideas have been used previously [83], where the notion of relatively coherent sets has been introduced. We note that our metastate subdivisioning process proceeds from the longest-lived states down rather than from the shortest-lived states up, where the latter is more common in temporal clustering approaches [84][85][86].…”