Using Machine Learning to Automate the Analysis of an Olfactory Habituation-Dishabituation Task in Mice

Area of Science:

• Neuroscience
• Behavioral Science
• Computational Biology

Background:

• Preclinical research relies on accurate mouse behavioral data.
• Current manual annotation methods are time-consuming and can be subjective.
• Improving data extraction efficiency enhances research throughput and scientific value.

Purpose of the Study:

• To develop and validate an automated pipeline for analyzing mouse performance in olfactory habituation-dishabituation tasks.
• To quantify odor interaction (sniffing time) using machine learning.
• To overcome challenges posed by occluded body parts in video analysis.

Main Methods:

• Developed an automated pipeline combining DeepLabCut for pose estimation and SimBA for behavioral classification.
• Utilized a single side-view camera for data acquisition.
• Trained machine learning models on manually annotated datasets and validated on unseen videos.

Main Results:

• The automated pipeline achieved high accuracy in estimating behavioral performance.
• Quantified odor interaction, including sniffing time, in a three-odor task variant.
• Machine learning-derived data produced comparable technical and biological results to manual scoring via linear mixed modeling.

Conclusions:

• The customized pipeline is validated for automated scoring of mouse sensory tasks.
• The automated approach enhances efficiency and accuracy in preclinical research.
• Discussed the strengths and limitations of the developed machine learning pipeline.