Project Objectives
- Develop advanced machine‑learning tools to analyse naturalistic rodent behaviour using 3D tracking.
- Demonstrate that these tools detect meaningful behavioural differences using fewer animals than standard tests.
- Build unsupervised deep‑learning models to identify subtle behavioural patterns not captured by traditional scoring.
- Package all algorithms into user‑friendly software modules and online tools compatible with DeepLabCut.
3Rs Impact
- Reduces animal numbers by increasing statistical power and lowering variability in behavioural studies.
- Replaces multiple stressful behavioural tests with a single homecage recording session.
- Refines welfare by enabling behavioural assessment in familiar environments rather than novel, stressful setups.
- Supports data sharing and re‑analysis, reducing the need to repeat experiments.
Background
Behavioural testing is central to neuroscience because it provides the most direct readout of brain function. Yet despite its importance, behavioural analysis in rodents remains limited by outdated tools, high variability, and superficial measurements. Commercial systems typically capture only basic metrics such as movement paths or time spent in specific zones, while human scoring—although more accurate—is slow, inconsistent, and limited to a small number of behaviours at a time. As a result, researchers often run large batteries of tests and use many animals to achieve statistical power, contributing to high animal use and poor reproducibility.
Recent advances in machine learning, particularly pose‑estimation tools like DeepLabCut, have transformed how movement can be tracked. However, these tools do not analyse behaviour themselves. This project aims to bridge that gap by developing supervised and unsupervised deep‑learning models that can recognise complex, ethologically relevant behaviours from 3D tracking data. By validating the approach in a mouse model of obsessive‑compulsive disorder, the team will show that deeper behavioural profiling can reduce group sizes while improving sensitivity.
The project ultimately seeks to modernise behavioural neuroscience, reduce animal use, and provide freely accessible tools for the global research community.
