Discrepancy between templates and actual waveforms

[teresajurado]

Hello everyone,
We have noticed that in 35% of the clusters sorted in our database, the waveform of the templates, as seen in Phy, does not match the average waveform calculated using the spike times in the raw data. Below, you can see three example clusters where this occurs. On the left is the screenshot of the cluster as seen in Phy, and on the right is the average waveform derived from the raw data.
Does anyone have any idea why this happens? Why is there such a significant difference between the two waveforms?
As can be seen in the AmplitudeView, these clusters do not show much drift that could explain the difference between waveforms. They are also not clusters resulting from merging.
Thank you!

[jacobpennington]

The templates are not always going to exactly match the raw waveforms. There also appears to be a channel offset of 1 in right-hand plots you're showing. I.e in your first example if you compare channel 18 in phy to channel 19 in your mean waveform plot, and channel 19 in phy to channel 20 in your plot, etc, they're actually quite similar. The middle example appears to be the most dissimilar, but it's also MUA.

It would probably be easier to see the differences if you toggle between template and waveform mode in Phy2's WaveformVIew instead of creating separate plots.

[teresajurado]

Thank you very much for your response, Jacob. Focusing only on the last example, differences can be observed between the waveform and the template. The template shows greater amplitude on channel 11, while the waveform shows it on channel 9. Why does the signal present in the template on channel 11 disappear in the waveform? Could it be due to drift?

For our experiments, it is important to identify the channel with the highest amplitude in the waveform because it indicates the actual location of the soma. Kilosort's output saves the templates, but is there any way to recover the waveforms or the channel with the highest amplitude in them?

Thank you.

[jacobpennington]

It's unlikely this is due to drift. This does look like a strange case: how common are examples like these in your data? Some other things that would be helpful to look at in Phy is the WaveformView option that shows many randomly selected waveforms instead of just the mean.

[teresajurado]

Thank you, Jacob,
We have calculated that 20% of our clusters change the channel of maximum amplitude of the waveforms compared to the channel of maximum amplitude of the templates. Attached is an image with waveforms of different numbers of spikes:
Waveform 1 (n spikes = 100), Waveform 2 (n spikes = 1000), and Waveform 3 (n spikes = 10,000).

[jacobpennington]

Sorry, I guess I didn't clarify that very well. What would be useful is a side-by-side comparison for the same unit. I.e. a screenshot of a single cluster showing template, average waveform, and raw waveforms. As for the examples you listed here, note that any cluster with 1000 spikes or less is likely to be noisy.