Machine learning-based automated phenotyping of inflammatory nocifensive behavior in mice

Area of Science:

• Neuroscience
• Pharmacology
• Biomedical Engineering

Background:

• Developing nonaddictive pain therapeutics requires efficient preclinical nociception assays.
• Current methods for scoring nocifensive behavior in models like the formalin assay are labor-intensive and variable.
• Automation of behavioral scoring is crucial for reducing costs and improving reliability in pain research.

Purpose of the Study:

• To develop and validate a machine learning-based automated system for scoring nocifensive behavior in the mouse formalin assay.
• To assess the accuracy and scalability of the automated system compared to human observers.
• To demonstrate the system's utility in identifying strain-specific differences in pain responses.

Main Methods:

• Utilized machine learning techniques on video recordings of mice undergoing the formalin assay.
• Implemented a three-component system: key point detection, per-frame feature extraction, and GentleBoost algorithm for behavior classification.
• Validated the system using 111 short videos (284 minutes) and longer experimental recordings (90 minutes) from two mouse strains.

Main Results:

• The automated system achieved 98% agreement with human observers in identifying hind paw licking/biting behavior.
• Successfully scored over 80 hours of video, demonstrating scalability for long-term experiments.
• Revealed significant strain differences in nociceptive response timing and amplitude between C57BL/6NJ and C57BL/6J mice.

Conclusions:

• The developed machine learning system offers an accurate, consistent, and user-friendly solution for automating nocifensive behavior scoring.
• This automation significantly reduces time and labor costs, enhancing the feasibility of the formalin assay for large-scale genetic studies.
• The system's reliability and scalability support the discovery and development of novel pain therapeutics.