Parameters for sorting recordings with tetrodes in cortex

[Stephan Boltzmann]

Hello everybody out there using Spyking-Circus,

It's nice to have Spyking-Circus available as open-source to the community and it's a pleasure to work with it.

I carefully tried to follow the documentation, particularly the explanation of the parameters on https://spyking-circus.readthedocs.io/en/latest/advanced/tuning.html

I had sorted a recording with tetrodes in the cortex using KlustaKwik, KiloSort and SpykingCircus: With default parameters, KlustaKwik gives me a number of well-isolated single units after manual curation with Klusters (and served as the reference for me, because I'm using it since quite a while). With some tuning of the parameters, KiloSort also gives me also a number of well-isolated single units in a shorter time.

The interface of Spyking-Circus is from my point of view the best of all three programs, but I fail to get a higher number of well-isolated single units for my tetrode recording in the cortex. This is presumably due to me failing to choose good parameters, because I can always find 2 or 3 well-isolated single-units in my recordings (see screenshot attached), but not more.

Therefore, I wanted to ask you which further diagnostics I could use in order to pin down the problem.

Kind regards

Stephan

[Pierre Yger]

Which version of the software are you using? What is your .params, and are you sure the data have been properly filtered? 

If you are using Windows, we recently patched a nasty bug that might alter the results. So be sure to use the latest version of the software. If you are still not satisfied, i.e. if you think you have not enough clusters compared to what you think you should have, try decrease merge_params to suppress local merges, and/or share with me the sanity plots produced by the software if you set make_plots = png in the [clustering] section

Best

Pierre

--
You received this message because you are subscribed to the Google Groups "SpyKING CIRCUS" group.
To unsubscribe from this group and stop receiving emails from it, send an email to spyking-circus-u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/spyking-circus-users/c843ad3f-1f10-4e76-a8f4-ef82554aac0bn%40googlegroups.com.

[Stephan Boltzmann]

Hi Pierre,

Thanks a lot for your answer. I'm sorry for not having provided more information:

I'm using the latest version from GitHub and tried it both on Linux and Windows.

I recorded data from 32 channels at a sampling rate of 20 000 Hz

My configuration file is attached to the mail:

I had increased the lower frequency for filtering and in the meantime tried some additional settings other than default.

The geometry of my probes looks like this (with increments of 200 for the second coordinate):

        'geometry':{
            0: [  0.0, 0.0 ],
            1: [ 20.0, 0.0 ],
            2: [ 40.0, 0.0 ],
            3: [ 60.0, 0.0 ],
        }

with radius set to 100.

Attached are some of the sanity plots generated during clustering - I'm afraid I'm lacking experience to interpret them properly.

[Stephan Boltzmann]

I'm making some progress: Decreasing the allowed variance of the amplitude for template matching helped me a lot.

My impression is that I get good putative principle cells (i.e. single units with low firing rates and clean refractory periods).

Nonetheless, I entirely fail to identify putative single neurons: There are no single units with a firing rate > 10 Hz which would have a clean refractory period.

I'll add again some diagnostic plots, which could help to pin down the problem.

[Stephan Boltzmann]

In the meantime, I also managed to identify cells with high firing rates.

I used default parameters with a slightly higher lower frequency for filtering (500 instead of 300 Hz).

Moreover, I found varying the amplitude limits for fitting to smaller values helpful for checking how well spikes are detected:

amp_limits = (0.7, 1.5)