Drift correction with low channel count

[arnabiswas]

We have a recording with 32 channel probes. Do you have recommendations for drift correction methods for such low channel count? Thanks!

[zm711]

The kilosort folks have recommended against drift correction in those cases. Maybe @cwindolf could give a small comment on drift correcting in the dredge context with lower density probes.

[cwindolf]

Hi, yeah, I can offer some thoughts. As @zm711 mentions, it can be challenging to work with lower density: first, localizing spikes can be hard if the density is too low; second, even if the density is high enough to localize spikes, the interpolation applied to the data to correct the motion may still have problems. Still, some 32-channel probes are short enough that their density could possibly work for both of these cases.

You might find Extended Data Table 1 in the DREDge paper (https://rdcu.be/ecqZf) useful for picking parameters for short probes, if your 32 channels are densely arranged. For instance, the Neuropixels Ultra probe there is ~300um long, so its parameters could be relevant. If you try it out and run into difficulties, I can try to help out with parameters, especially if you are able to share visualizations (of spike locations over time with the motion trace overlaid, for instance, as shown here: https://spikeinterface.readthedocs.io/en/stable/how_to/handle_drift.html)

[arnabiswas]

Thanks @cwindolf we have a linear 32 channel probe with 25 micro meter spacing spanning 800 micrometers - that seems to be approximately as dense as the nueropixels 1.0 - I'll try both 1.0 and NP ultra params and get back. Thanks!

[arnabiswas]

Hi @cwindolf Here are a couple outputs below. The peak detection does not look great to me. Or do you think this is okay? We looked at the recordings in phy after motion correction and some units did seem decent. What is your recommendation, do we try motion correction or give it up altogether as bad motion correction may be worse than no motion correction?

Thanks!

The one below uses the NP Ultra parameters you pointed me to

This one below was run by my colleague runs the default dredge spikeinterface parameters.

[cwindolf]

Hey @arnabiswas, yeah, those detected spikes would not encourage me to be faithful in the dredge output from this pipeline, since it's not easy to visually guess what the motion should be. If KS is working, do you happen to have a driftmap image from KS? Or are you running it without motion correction? (If so, are you doing any post-hoc curation to account for the drift?)