🧠 UnitRefine Mice SUA Classifier

📌 Model Summary

This model is part of the UnitRefine pipeline and is trained to classify single-unit activity (SUA) in mouse Neuropixels recordings. It uses supervised machine learning to distinguish well-isolated units from multi-unit activity (MUA) and noise based on unit-level spike metrics.

The classifier is designed for fast, automated unit curation, and generalizes across multiple recordings and brain regions, achieving high accuracy even with limited training data.

The training data includes recordings from the Allen Institute for Neural Dynamics, International Brain Laboratory, and Musall Lab.

🔍 Use Cases

Automated post-processing of spike sorting output
Removing low-quality or noisy units prior to analysis
Reducing manual curation effort in large-scale neural recordings
Benchmarking unit quality metrics against expert annotations

🧬 Metric Selection

For information on which spike metrics were used to train this classifier, please refer to the model_info.json file included in the repository.

💡 How to Use

This model can be used to automatically identify SUA units from spike-sorted data. If you are working with a SortingAnalyzer object, you can run the following:

from spikeinterface.curation import auto_label_units

labels = auto_label_units(
    sorting_analyzer=sorting_analyzer,
    repo_id="AnoushkaJain3/UnitRefine-mice-sua-classifier",
    trusted=["numpy.dtype"]
)

This returns a dictionary of predicted labels per unit (1 = SUA, 0 = MUA/Noise).

📜 Citation

If you find UnitRefine models useful in your research, please cite: biorxiv paper.

🔗 Resources

GitHub Repository: UnitRefine
📖 SpikeInterface Tutorial – Automated Curation:
View Here

UnitRefine is fully integrated with SpikeInterface, making it easy to incorporate into existing workflows. 🚀

🙏 Acknowledgments

Special thanks to Alessio Buccino, Olivier Winter, and Alejandro Pan-Vazquez for generously providing the datasets used to train and evaluate this model.

👩‍🔬 Authors

Anoushka Jain
PhD Researcher, Musall Lab, Forschungszentrum Jülich

Chris Halcrow
Lead Developer, SpikeInterface