Weird line in feature view in phy

[miralabdalaziz]

Hi all,

I hope this is the right place to ask these questions, if not, let me know where I should post these.

We have a couple of questions about curating units in phy after running kilosort4.
We are using a dense neuronexus 64 channel probe (V1x64 Poly 3). Channel spacing is 25 micrometers and the recording sites span 550 micrometers in total.

1. When looking at single isolated clusters, sometimes we get a weird line in the feature view of the highest amplitude channel for that cluster. Why? What does that mean? Attached are two examples (Unit 87, Unit 298).
2. When looking at two separate clusters together, depending on the channel feature view, sometimes we get a line. What does that mean? (Unit 283 and Unit 297; ch 10 view shows a line while 12 doesn't).
3. When looking at two separate clusters together, depending on the channel feature view, one view tells me I should merge while the other doesn’t. What’s the best way to make decisions in these instances? (Unit 118 and Unit 108; ch 41 tells me I should merge while ch 57 shows a separation).
4. For the amplitude view, there's a feature based option and template based option. The feature based one is different when clicking on different channels which makes sense while the template one is not. What does that mean?

[jacobpennington]

1 & 2) That just means the values along one of the feature dimensions is very small relative to the scale of the plot, there's no specific interpretation for it. You'll see that often for MUA units (which all of your examples are).

3. In general, if you're making split/merge decisions based on feature view it's not necessary for every channel to show separation. However, in this case, you're comparing two units that are already marked MUA and likely have multiple units lumped together, so merging those would just lump even more units together. You should instead try to split each of those separately if you can (may not be possible).

4. That would be a question for the Phy github.