DeepOF

Lucas Miranda and colleagues at the Max Planck Institute of Psychiatry have developed DeepOF (Deep Open Field), a powerful open-source software package designed to process pose estimation data acquired using tools like DeepLabCut or SLEAP. DeepOF’s transforms raw video footage into structured time-series data of tracked body parts, enabling sophisticated post-hoc analysis, embedding of motion tracking data, and automated behavior annotation.

A major strength of DeepOF is its comprehensive approach to annotating and interpreting behavioral patterns, offering both established and novel methods:

Supervised Pipeline: This method uses rule-based annotators and pre-trained machine learning classifiers to reliably detect and measure predefined behavioral motifs (patterns you already know you’re looking for).

Unsupervised Pipeline: This advanced feature uses deep clustering models to automatically extract and identify novel behavioral motifs without any prior definition.

DeepOF is one of the first software packages to seamlessly integrate both supervised and unsupervised annotation pipelines. To help researchers understand the results from the unsupervised clusters, the software includes an interpretability pipeline using SHAP (SHapley Additive exPlanations) to explore the behavioral meaning of the extracted motifs.

This accessible and robust software is available for download on the lab’s GitHub, along with links to the required DeepLabCut and SLEAP software.

Thanks to Ellie Blumer for writing this post. She is a second-year neuroscience major and computer science minor at American University. She will be assisting with website content this year.