Automated phenotyping of rodent behavior in the Cylinder Exploration Test using machine learning.

Area of Science:

• Neuroscience
• Behavioral Science
• Computational Biology

Background:

• Rodent models are crucial for neuroscience research, requiring precise behavioral characterization for functional readouts.
• Accurate behavioral phenotyping is vital for understanding genotype-phenotype relationships and translational research.

Purpose of the Study:

• To develop a machine learning framework for automated rodent behavior analysis using the Cylinder Exploration Test (CET).
• To quantify specific behaviors like freezing, rearing, and locomotion.
• To identify key features differentiating experimental conditions and genotypes.

Main Methods:

• Utilized pose estimation and explainable machine learning for automated behavior analysis in the CET.
• Developed a framework to quantify freezing, rearing, exploratory movement, and locomotion.
• Validated the approach by phenotyping dopamine transporter knockout (DAT-KO) and tryptophan hydroxylase 2 knockout (Tph2-KO) rat strains and their wild-type controls.

Main Results:

• The automated framework successfully identified distinct strain-specific behavioral phenotypes.
• Achieved high classification accuracy: AUC=0.84 for DAT-KO vs. DAT-WT and AUC=0.98 for Tph2-KO vs. Tph2-WT.
• Characterized discriminative behavioral features between different genotypes.

Conclusions:

• Automated CET analysis can detect genotype-specific behavioral signatures in rodents.
• This scalable method supports standardized phenotyping in neuroscience and preclinical research.
• The framework offers a robust tool for analyzing complex behaviors and their genetic underpinnings.