Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Drifting issue #43

Open
DCampa opened this issue Dec 15, 2016 · 1 comment
Open

Drifting issue #43

DCampa opened this issue Dec 15, 2016 · 1 comment

Comments

@DCampa
Copy link

DCampa commented Dec 15, 2016

Hello,
I am spike sorting data collected with an O3 Neuropixel probe during acute experiment.

During some of my recording sessions the brain tissue moved in respect to the probe.
The outcome is that during the recording you observe a variation in the size of the spikes fired by a given neuron and in which recording sites they are.

Often what Kilosort does in this situation is simply to split the same unit into multiple clusters over time, and all I have to do is to merge them.

However there are situations in which which clusters belong to the same unit is not so obvious or there are so many units in which the drift occurred that the manual merging becomes time consuming. I was wondering whether Kilosort has an automatic way to correct for this problem or if there is a combination of initial parameters which helps with this.

Thank you
Dario

@nsteinme
Copy link
Collaborator

Hi Dario - one parameter you can play with is the number of templates you ask to be fit (ops.Nfilt). If you make this smaller, Kilosort will do the best it can with that restriction and so it'll end up being more inclusive about which spikes get included in each template, i.e. it should do some of the merging for you. However there is a tradeoff in that you will also get more instances in which spikes from separate neurons are incorrectly combined into a single template, requiring you to split them out. Since splitting takes longer and is more error-prone than merging, the general recommendation is to err on the side of doing more merging.

More generally, of course everyone's recordings do have drift including ours so we are definitely aware of the problem. The Neuropixels probes potentially offer a great opportunity to overcome this computationally since the same neurons are usually still observed by some other sites when the brain moves. Marius has been working on an implementation of a post-hoc drift correction, and he'll have to provide you with more details, but the short answer is we don't have any working solution for it yet.

Nick

lshaheen pushed a commit to LBHB/KiloSort that referenced this issue Mar 17, 2021
add graph labels for diagnostic plot
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants