Automation of mouse wipe behavior in animal behavioral studies using machine learning

Authors

  • J Semseddin
  • G Salem
  • M Garmendia
  • S Arumugam
  • H McNabb
  • N Mokhasi
  • J Zhang
  • T Pohida
  • WF Simonds

Abstract

Conditional KO of the Gnb5 gene in sensory neurons using Advillin-Cre+/-; Gnb5flox/flox mice demonstrated that Gβ5, the protein encoded by Gnb5, is a key regulator of pain sensation. The eye wipe assay was used to quantify mouse nociception via the trigeminal sensory nerve; drops of dilute mustard oil-isothiocyanate or capsaicin were applied to the cornea and the number of eye wipes was manually counted for one minute. To facilitate data collection and improve accuracy, we developed an apparatus to automate this task through machine learning. Ensuring high-quality video datasets necessitated the design/assembly of a custom-made apparatus consisting of both hardware and software. The hardware contains a well-lit triangular arena surrounded by mirrors to allow capture of different views of the mouse and a camera to record the entire arena. This setup results in a small footprint system convenient for the laboratory setting. One-minute videos were taken of each mouse with the irritant delivered into either its left or right eye. After dataset compilation, we utilized an AI platform (Encord) to annotate videos of the mouse eye wipe frame-by-frame, review for errors, and feed the labeled data into training machine learning models to automate eye wipe classification and quantification. AI integration into our apparatus is expected to increase the accuracy of the eye wipe assay, significantly reduce observer bias/human error, and adaptable to analyze additional behaviors, freeing up time and resources. This system will be a great asset to laboratories looking to maximize the quality, reproducibility, and efficiency of behavioral experiments.

Scientific Focus Area: Neuroscience

This page was last updated on Tuesday, August 6, 2024