Syntactic Grooming Classification in Mice with Computer Vision

Cayson Hamilton

Document Type

Article

Publication Date

8-9-2024

Keywords

JMG

JAX Location

In: Student Reports, Summer 2024, The Jackson Laboratory

Abstract

Behavior analysis is a challenging yet elucidating part of disease research. Specifically behaviors such as grooming, which models many stereotyped, repetitive, or comforting behaviors in humans, can provide much insight when studying models of psychiatric disease in mice – Autism Spectrum Disorder or Obsessive Compulsive Behavior for example. However, grooming is only a broad behavior label for a set of more intricate grooming types, or syntaxes. It has been observed that these grooming syntaxes are typically linked together in certain orders to form syntactic chains. A mouse’s inclination to follow said syntactic chains or lack thereof can be especially elucidating, as it can tell us alot about the physical and psychiatric state of the animal, as well as if it may be affected by environmental conditions such as stress. The challenge with using this behavior remains that these syntaxes are difficult and inefficient to manually annotate. This limits the duration and setting in which these metrics can be analyzed. Here, we provide an automated solution using computer vision to classify grooming syntaxes and identify syntactic chains across a wide variety of mice strains. Our multiclass classifier provides an efficient way of processing high-throughput data, enabling more comprehensive interventional studies concerning diseases with related behavioral phenotypes. We compare syntax distributions across a strain-survey containing 62 distinct mouse strains. Further, we apply the field’s standard protocol to investigate mice’s inclination towards a pre-defined syntactic chain, and discover that the established chain definitions fail to account for the vast majority of grooming bouts.

Please contact the Joan Staats Library for information regarding this document.

COinS