Automated clustering of larval zebrafish motor behavior reveals two different modes of fast escapes.

AGULLÓ, VALENTÍN; MARTORELL, NICOLÁS; MEDAN, VIOLETA

San Luis

Congreso; SAN2023; 2023

Sociedad Argentina de Neurociencias

Detecting threats and triggering adequate evasive behaviors when warranted is one of the most critical behaviors animals perform. However, not all stimuli trigger the same evasive behaviors and we still have a rudimentary understanding of how the brain processes sensory information to select the correct motor behavior.Zebrafish show a diversity of evasive behaviors some of which have a known neural basis, but how stimulus salience affects the selection of specific motor behavior is not known.To bridge this gap, we filmed behavior of larval zebrafish while presenting a set of threatening visual and auditory stimuli, which triggered diverse evasive responses. Since manual labeling of hundreds of individual responses is extremely time consuming and prone to biases, we developed an automated pipeline to identify fast evasive behaviors.We first extracted animal trajectories using DeepLabCut. Next, we developed an algorithm that segments the trajectory and recognizes rapid motor events, including evasive responses. Employing methods for dimensionality reduction (t-SNE, UMAP, autoencoders) and clustering (K-means, Random Forest), we categorized the events into three consistent groups: 1) slow escapes to visual stimuli, 2) fast escapes to auditory and multisensory stimuli, 3) non-evasive fast reactions. Our findings strongly support the hypothesis of two discrete modes of escape: fast and slow with no intermediate motor patterns.