Steinmetz et al, 2019 dataset questions

Yes, those are miss trials, and a very few of them are cut off because they’d be very long.

1 Like

Hi guys, I was wondering how the wheel movement per trial from the example dataset provided here: colab notebook was calculated from the raw data from the figshare repository?

@pachitarium :slight_smile:

You can see that in the script and function I linked above (here). The wheel trace is a cumulative encoder, so you have to take differences first. In this case, the differences were sampled at 10ms, same as the sampling for the other variables.

Thanks for your answer.

I don’t see this in the curated data. Are you talking about amps.npy? I want to compute the power spectrum.

Make sure you are using the latest data loader from here. The LFP is one of the many variables included, here’s a screenshot from the notebook:

I see what is the confusion. We were looking at this
Considering that maybe the sample data was not going to be enough. Is it not possible we can have access to the scripts that go from the raw data to the curated data?


I was looking through the Steinmetz dataset and noticed that in the ‘spks’ parameter (dat[‘spks’]) there were a couple of different potential values at each of the time bins (Values could be 0,1,2,3,4 etc.). I was initially under the impression that 0 was no neurons firing and 1 was neurons firing. Can someone explain to me what these values mean? Could it be number of neurons firing ?

The scripts are available, read above in the thread to see where they are.

That’s the number of spikes that neuron firing in that particular 10ms bin.

We were also using the neuromatch loader in the NMA notebook for Steinmetz dataset. Unfortunately it seems to be in 10 ms bins, which is not good enough to estimate coherence of gamma oscillations for our project. Is there a way to get less downsampled LFP data?

For consistency, everything was binned at 10ms (gamma is around 50Hz, right?). If you want smaller binning, you can re-run the processing on the raw data. I have shared scripts above to do that.

My two cents: don’t focus too much on the LFP. The strength of this dataset is the large number of single-units distributed across the entire brain.

1 Like

For 3. Does response = 0 only correspond to no wheel movement despite incentive?
This may occur for either condition right? So, either when equal right/left contrast exists or for incentive/unequal contrasts (which implies a failed trial). So response = 0 would mean the mouse did not move with or without incentive?

1 Like

Interesting! Any way to extract data for when the movements prior to the go cue (sound cue) are not in the same direction as the detected response direction?
Also, the few times when this did happen, was it more likely to produce an incorrect response?

@pachitarium My apoligies for so many questions, but are you still on adding the spike times?
I am trying to get it using the script but I am failing terribly to understand how the spike times are being reshaped there. (from a flat array to spikes for each neurons) Any suggestions?

1 Like

Alright, thanks Pachitarium.

@h.rajpal I am still doing that, will be added today together with other updates.
@sanjoshi897 Response = 0 can happen in any condition. Equal, non-zero contrast actually doesn’t have a correct response (it’s rewarded probabilistically, 50/50 for either movement).
@sanjoshi897 I also want to extract the times of the first movements from the wheel data, also due today. There really aren’t many trials when the initial movement is opposite to the response, but it could be interesting to consider the full wheel dynamics over time, with starts and stops and increases in vigor etc.


@pachitarium Just want to confirm, if the wheel is the position, not the velocity? Is the wheel’s position locked with the horizontal translation of the stimulus? Before the go cue, does the animal keep spinning it or somehow just turn a bit, then stop, and then reset to original position?

@pachitarium Hi, I am interested in the randomness of the delay interval after the stimulus onset—what is the reason for this as opposed to always having a fixed interval before the auditory cue?

In the curated dataset, the wheel is the speed (on figshare its the cumulative ticks of the rotary encoder). The wheel is locked to the stimulus only after the go cue. Before that the stimulus is fixed. Before the go cue the mouse does keep spinning the wheel, almost always in the same direction as the one it eventually reports after the go cue.

1 Like

Its to keep the mice on their toes! Nick didn’t want the mice to get super stereotyped in their actions, because there is a worry that overly stereotyped behaviors don’t really engage the brain in the same way as more purposeful behaviors.