search

Julie-Fabre / bombcell

Automated quality control, curation and neuron classification of spike-sorted electrophysiology data
GNU General Public License v3.0
123 stars 30 forks source link

No units - change params? #178

Closed ganupuru closed 1 week ago

ganupuru commented 2 weeks ago

Hello Julie, I'm trying to use bombcell to curate a Neuropixels dataset sorted with Kilosort 4. I have 587 clusters in this set, and looking at them in Phy2, I can pretty confidently see that there is real neural activity there, whether or not they're good units.

However, bombcell is not picking up anything (no good units or MUA), and as you know, it recommends changing parameters. Unfortunately, I'm not savvy enough to know where to start. Are there any obvious ones to tweak first, or specific metrics that might indicate?

Preetham

Julie-Fabre commented 2 weeks ago

Hi Preetham,

Thanks for your message!

It does sound like you'll need to tweak some parameters. I'm curious about why things are so different for you. Could you tell me which brain regions you're recording from, in which animal model, and in which setting (headfixed acute, headfixed chronic, freely-moving, ...).

Could you send me the output figures (just a screen shot is fine) of bombcell? The plots that look something like this (below) image image image image image

ganupuru commented 2 weeks ago

Thanks for the quick reply. I'm recording from rat prefrontal cortex, chronic/freely moving. Here are my outputs:

bombcell_output

Julie-Fabre commented 1 week ago

Hi Preetham,

Thank you for the information and the plots! Are you sure this is the correct waveform plot? There are very few waveforms plotted despite many neurons in the UpSet plots (the blue histograms).

ganupuru commented 1 week ago

Yes, that's the one that's displayed along with those other plots.

Could this be an issue with how I'm loading the data? If it makes any difference, I'm defining ephysRawFile as the continuous.dat file that outputs from open ephys (the file itself, not the directory). I'm defining ephysMetaDir as a structure (not a character array) where my oebin file is. I'm defining the ephysKilosortPath as the "kilosort4" subdirectory.

Julie-Fabre commented 1 week ago

Sorry I think I got confused by your top MUA upset plot (which suggested you had more than 400 MUA units)- I guess the bottom one is the correct one? :)

In that case, the proportions don't look so off to me - 2/18 (>10 % are good single units - we usually expect this number to be ballpark around between 10 to 25 %) . That being said, looking at the histograms, there are some value you can tweak but I'm not sure the histograms are the ones for your dataset though. I can see you have a slightly older version of bombcell too (from the plots). Would you be able to download the latest version of bombcell are re-run it on your dataset and then add the output histogram again? Thanks a lot!

image

ganupuru commented 1 week ago

OK, reinstalling bombcell completely fresh, I'm now getting the following error in MATLAB at line 130 of extractRawWaveformsFast: Unable to perform assignment because the size of the left side is 1-by-61-by-383 and the size of the right side is 383-by-61.

Which seems like a pretty minor error, but it looks like one of those variables, rawWaveForms or rawWaveformsFull, is shaped incorrectly. Is this error immediately familiar to you, or should I work backwards through the code and figure it out? Sorry, I know this is outside the original issue.

Julie-Fabre commented 1 week ago

No need to apologize, thank you for bringing that up! A small bug had crept in (sorry about that!). Could you pull again? Things should be fixed now :)

ganupuru commented 1 week ago

OK, I think we're good now. I'm getting sensible results after updating. Thanks for all the help.

Julie-Fabre commented 1 week ago

Great! Let me know if you run into any other issues or questions :)