How to compare spike sorting runs with no ground-truth - Upload IBL benchmarks datasets in Spike Forest
Key Investigators
- Olivier Winter (IBL)
- Kush Banga (IBL)
- Mayo Faulkner (IBL)
- Jeremy Magland (Flat Iron Institute) - For guidance
Project Description
We have a set of 4 Neuropixel benchmark datasets that we use at IBL to compare spike sorting algorithms and parametrization in different brain regions. It may be a good thing to share those 4 datasets more broadly, possibly in spike forest for others to use them as benchmarks. There is no ground truth for this data.
Objectives
The objectives are 2-fold: get the IBL people acquainted with spike forest by contributing datasets, and a larger objective that is to try define metrics for spike sorting runs when no reference / ground truth is available.
Approach and Plan
1) See if it is desirable to upload those datasets on spike forest, if yes, do it. 2) Run a few sorters on those datasets (otherwise there are already some results available from IBL) 3) Devise some functions / code to compare and assess the quality of the runs.