OpenEphys contact_ids inconsistent with SpikeGLX when `SELECTED_ELECTRODES` is not available

[h-mayorquin]

The read_openephys() function computes elec_ids from electrode positions when the SELECTED_ELECTRODES field is not available (the common case for older OpenEphys recordings). This reverse calculation (position to elec_id) produces results inconsistent with SpikeGLX when the positions stored by OpenEphys do not match the canonical probe geometry. (see @chrishalcrow in #383 (comment)). This inconsistency means the same physical electrode has different contact_ids depending on whether the recording was made with SpikeGLX or OpenEphys. I think Neuropixels probes should have the same contact_ids regardless of which recording system was used.

After #383 is merged, we should add consistency tests up to the point of building the full probe so they should match if possible. If they don't match, we should make this more explicit in the code and document it. A potential fix would be to use build_neuropixels_probe() to create a full probe, then match recorded positions to the canonical electrode grid using nearest-neighbor lookup, ensuring consistency with SpikeGLX regardless of what positions OpenEphys stores.

[alejoe91]

Thanks for writing this up @h-mayorquin

The Open Ephys team is aware about this and that's why we decided to switch to the SELECTED_ELECTRODES field for new versions.
I think it's a good idea for us to have a consistency test between SpikeGLX and Open Ephys, but I also think that the important thing is that relative positions are the same (which should be!). Contact IDs are a name that we arbitrarily assign to contacts, so strictly speaking they are not required to be the same between openephys and SpikeGLX. Happy to discuss more about how to properly deal with these discrepancies

[chrishalcrow]

Yeah, would be good to talk in person. Just to note: we use contact ids in our lab. We keep a csv with all contact ids alongside their anatomical coordinates. So although they are arbitrarily assigned by us, I'd bet a few people do use them. So I'd prioritize backwards compatibility over consistency between openephys and spikeglx.

[h-mayorquin]

Yes, I would prioritize consistency. I think to the degree that is possible the library should be acquisition system independent and have canonical ids that can be mapped across probes instances. Of course, if that is not possible or too costly we should just document it better as mentioned above.