EventDetection not appropriate for modern probes

[bendichter]

The EventDetection class in the ecephys module is meant to store the detected unsorted spike times, e.g. threshold crossings of an ElectricalSeries. This type of object is rarely if ever used because users generally prefer to store the raw and/or output of spike sorting, and have little use to store intermediate values in the spike sorting pipeline. Nevertheless, I was attempting to write a section of our ecephys tutorial for the type to increase our neurodata type coverage in the documentation. Upon doing so I realized there were some oddities in this ndtype.

First of all, it requires both the source_idx of the spike in the source ElectricalSeries and the timestamp, which is redundant.

Second, it does not contain information about which electrode(s) detected the spike. This might be acceptable with probes that have a small number of electrodes, such as tetrodes, but for modern probes like Neuropixels that have 100+ channels, I think this is of limited utility without indicating which electrodes are involved in the spike.

I'm not sure how to proceed here. Has anyone seen people actually use this type? Should we fix it or deprecate it?

[CodyCBakerPhD]

Has anyone seen people actually use this type?

One way to help answer this question, which can be used for justification of its removal, would be to scan all files on the DANDI archive to see if any of them contain a single object of this type

If not, I say deprecate. We use the Units table for spike thresholding all the time, with the caveat that we usually put it under a processing module instead of the outer level

[stephprince]

I just noticed this open issue and it relates to upcoming schema changes so wanted to add to the discussion here.

First of all, it requires both the source_idx of the spike in the source ElectricalSeries and the timestamp, which is redundant.

Second, it does not contain information about which electrode(s) detected the spike. This might be acceptable with probes that have a small number of electrodes, such as tetrodes, but for modern probes like Neuropixels that have 100+ channels, I think this is of limited utility without indicating which electrodes are involved in the spike.

These issues are addressed in #620 and NeurodataWithoutBorders/pynwb#2091 by making timestamps optional and adding the option to specify source_index as either just the timestamp index or the timestamp and the channel indices.

Has anyone seen people actually use this type? Should we fix it or deprecate it?

I didn't see any examples on DANDI, but these changes were in response to a user issue (NeurodataWithoutBorders/pynwb#2068) where they had a micro electrode array in a cell culture well plate and wanted to use EventDetection.

We use the Units table for spike thresholding all the time, with the caveat that we usually put it under a processing module instead of the outer level

How does the Units table for spike thresholding work - is a unit all spike events from a single channel? I'm wondering from this discussion if we want to keep EventDetection with the upcoming schema modifications or recommend a different data type instead.