An important concept in ethology is that complex behaviors can be constructed from a set of basic motor patterns. Identifying the set of patterns available to an animal is key to making quantitative descriptions of behavior that reflect the underlying motor system organization. We addressed these questions in zebrafish larvae, which swim in bouts that are naturally segmented in time. We developed a robust and general purpose clustering method (clusterdv) to ensure accurate identification of movement clusters and applied it to a dataset consisting of millions of swim bouts, captured at high temporal resolution from a comprehensive set of behavioral contexts. We identified a set of thirteen basic swimming patterns that are used flexibly in various combinations across different behavioral contexts and show that this classification can be used to dissect the sensorimotor transformations underlying larval social behavior and hunting. Furthermore, using the same approach at different levels in the behavioral hierarchy, we show that the set of swim bouts are themselves constructed from a basic set of tail movements and that bouts are executed in sequences specific to different behaviors.
Keywords: behavior; behavioral motifs; cluster analysis; clusterdv; locomotion; motor control; sequences; unsupervised machine learning; visual behavior; zebrafish.
Copyright © 2017 Elsevier Ltd. All rights reserved.