Back

UnitRefine: A Community Toolbox for Automated Spike Sorting Curation

Jain, A.; Greene, R.; Halcrow, C.; Swann, J.; Kleinjohann, A.; Spurio, F.; Graff, S.; Kampa, B.; Gall, J.; Gruen, S.; Buccino, A. P.; Hennig, M. H.; Musall, S.

2025-04-04 neuroscience
10.1101/2025.03.30.645770 bioRxiv
Show abstract

High-density electrophysiology simultaneously captures the activity from hundreds of neurons, but isolating single-unit activity still relies on slow and subjective manual curation. As datasets keep increasing, this poses a major bottleneck in the field. We therefore developed UnitRefine, a classification toolbox that automates curation by training various machine-learning models directly on human expert annotations. Fully integrated in the SpikeInterface ecosystem, UnitRefine combines established and novel quality metrics, cascading classification and comprehensive hyperparameter search to provide optimized models for different applications. UnitRefine achieves human-level performance across diverse datasets, spanning species, probe types, and laboratories, including recordings from mice, rats, mole rats, primates, and human patients. Applied to a large brain-wide dataset, UnitRefine doubled single unit yield and improved behavioral decoding performance. A streamlined graphical interface allows models to be fine-tuned to new datasets and shared via the Hugging Face Hub, enabling broad adoption and community-driven improvement of automated curation workflows.

Matching journals

The top 4 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.