unmatching sorted units and distorted waveforms when sorting merged recordings

[mdandola]

Hello,

I am trying to track the activity of single units across different recordings of the same implant of NP1 (phase 3A). As for the October's biorXive preprint about NP2.0, I tried to do a single sorting with kilosort3 after merging 2 recordings taken in 2 different moments. The same 2 recordings were sorted previously separately using kilosort2 and I compared the results for crosschecking. I noticed that the units identified when sorting the two merged recordings are only partially overlapping with the ones obtained when sorting the recordings separately. In particular, the merged sorted output was full of what looked like noise clusters pre-classified as good, the number of MUA clusters was 10 times higher, and the few clusters that I could somehow match with the ones obtained when sorting separately have distorted waveforms (which made me think that also some of the cluster that I classified as noise could actually be good with some sort of distorting in waveform).
For example, please look at the two screenshot that I took as an example: they show one of the units I could partially match when sorted separately and after merging.
What would you suggest me to do? Do you think this is a matter of adjusting sorting parameters, or am I doing something wrong?
A couple of details:

• the merging was done using the preprocessing toolbox of spike-interface
• the two recordings were taken from a fixed spinal cord preparation a few hours apart
• sorting parameters were the same in the two cases

thank you for your help!
Mattia

Separated sorting:

Merged sorting:

[mdandola]

Hi, does anyone have some suggestions about this? the issue is been open for 15 days without comment