BUG: <no spikes detected with 4*4 distand probes>

[liqun2017]

Describe the issue:

Hi,

I am a beginner of kilosort. I would like ask how to set the parameters for kilosort4 with 44 distand electrodes? I made a probe map with 44 (200 microns each) (Fig.1). The data seems to be loaded correctly (Fig. 2), but no spikes detected. I have checked the similar issuses which been brought up, and lowered nearest templates to 50, increased dminx to 200 (Fig. 3). I have awared that the kilosort is designed for high density probes, while it seems can work with tetrodes (people said). I would like kindly ask if it can work with my situtaions? If it is so, how can I define parameters to get the results?

The kilosort.log file is uploaded here:
kilosort4.log

Fig.1

Fig.2

Fig.3

I would be grateful for your response.

Best,
Qun

Reproduce the bug:

No response

Error message:

No response

Version information:

python 3.9.20
kilosort v4.0.20
window 11

[jacobpennington]

Can you please check the boxes for "universal templates" and "grouping centers" under the probe preview in the GUI and upload a screenshot of what that looks like?

[liqun2017]

Sure, please check if there is anything wrong?

[liqun2017]

Hi, I would like ask if this probe map I made is correct or not? Since I realized the y-axis may indicate the depth of recording site located in the brain. However, our probe is made of 4 * 4 electrodes with the only one recording site in the tip of each shank, it's more like a plane at the same depth in the brain, should I use some kind of 1 * 16 probe map to indicate this situation, or it will not affect much?

[liqun2017]

Sorry for many questions. I went through the docs about "When to adjust default settings" carefully. Then tried to set paramters as follows, seems some results came out, could you please have a look at the parameters if there are good?

Red lines mean the parameters have been modified accroding to those who are metioned in the docs

[jacobpennington]

@liqun2017 With the spacing of your contacts, a 1x16 probe map would probably work better (i.e. 16 channels along the vertical axis, all with the same x coordinate). The y-axis is indeed interpreted as depth in the brain, but as long as drift correction is disabled that isn't a problem.

I would recommend creating a new probe map for that and sort with default settings again (except for fs, nt, nblocks, and batch_size, those seem appropriate). In particular, setting Th_universal and Th_learned to very small values is not likely to be helpful. You're probably just getting spikes detected now because it's incorrectly detecting small noise fluctuations as spikes.

[liqun2017]

I see, thank you so much. I will redo it carefully again.