Marigold

Marigold analyzes zebrafish behavior through machine learning-based pose tracking of up to 10 user-defined keypoints and export of kinematic parameters for high-throughput behavioral studies.

Key Features:

• Custom neural network architecture: A highly efficient, custom-designed neural network achieves reasonable training and inference speeds on modestly powered computers lacking a discrete GPU.
• Multiwell format support: The tool supports both single- and multiwell recording formats for high-throughput experimental designs.
• Kinematic parameter export: Marigold exports a range of kinematic parameters alongside publication-quality data visualizations.
• Cross-platform web deployment: The tool runs within popular web browsers on ChromeOS, Linux, macOS, and Windows without installation.
• Dynamic background compatibility: The pose tracking functions with recordings containing dynamic backgrounds or foreign objects.

Scientific Applications:

• Neurological disorder modeling: Behavioral analysis supports studies of autism, epilepsy, and other neurological disorders through zebrafish models with human disease gene orthologs.
• Drug discovery and toxicology: High-throughput behavioral phenotyping facilitates screening applications in pharmacological and toxicological research.
• Glutamate transporter mutant analysis: The tool enables examination of touch-evoked escape responses in tnt mutant embryos harboring loss-of-function mutations in Eaat2b.
• Visuomotor behavior studies: Marigold quantifies larval swimming bout number and vigor under different feeding regimens at 5 and 7 dpf.

Methodology:

Marigold uses a custom-designed neural network architecture for pose tracking of user-defined keypoints in video recordings of embryonic and larval zebrafish.