Steinmetz et al, 2019 dataset questions

Hi everyone! My students asked me what I thought were quite reasonable questions about this data set so I was hoping maybe others could point them/me to the following data:

  1. Do we have data from the ITI periods? (Having looked at the datasets noteboook I also struggled to find the timestamps of different events - should be made clearer (if they even exist) in my opinion so I understand their frustration).
  2. Are trial times aligned or are there timestamps (e.g. at what point a sound signal is given in each trial)? (again goes back to the first question - why are there 250 time points, 10ms binned so 2.5s length of trials which again doesn’t seem to be consistent with the paper methods so at what points does every event happen?)
  3. Are there any recordings of those trials when the mouse did not move, despite the fact that there is an incentive?
  4. Is there any information about whether a trial was rewarded or not? (for example, in those trials where reward probability is = 50/50).

I looked at the notebook link and explored the data a bit and couldn’t find any of this information either. Could someone who has had more experience with this data set clarify these points for us? Thank you :slight_smile:


Nick is on here, so I’m tagging him @nsteinme. However, I put the dataset into that format, which has advantages and disadvantages, but it was primarily designed to make the trial structure simple and easy to align variables.

  1. Other than the 500ms of pre-stimulus activity: no. However, there isn’t really much more data in the ITI, the mice are pretty fast.
  2. Aligned to stim onset, from -500ms to 2000ms, regardless of other trial timestamps. The timestemps provided right now are stim_onset (always 500ms, and response_time: when the mouse’s choice was registered after the go cue).
  3. These are automatically included and have response = 0.
  4. I will add that information, together with reward times.

I have a technical question regarding the paper. Some neurons in VIS responded to action, but was there a control for the motion signal induced by the action? @nsteinme

Sorry I should probably just read the paper, but I find myself out-of-memory and out-of-time.

The closest thing would be the passive trials, which only had stim but no behavior. I have not included those, but I think I might be able to.

Isn’t the action kernel restricted to a window before the go cue (such that the stimulus is fixed)?

There is actually a lot of movement before the go cue, which happens 500-1200ms after stim onset. The mouse however starts turning the wheel at around 200-250ms after stim onset.

1 Like

Francisco is almost right but to clarify: the action kernel is restricted to the time of earliest detectable movement onset. Kevin I don’t think I understand what you mean by “motion signal induced by the action” - what’s “motion signal”?

1 Like

are there recordings made during training sessions of the task (when the animal is still learning how to do the task?)


Hi, yes sorry the dataset as it is arranged at the moment definitely has many advantages. One of my groups so far did not need any of these things as they mostly want to work on visual processing but the other group wants to work on decision making so these things might be important for them.

If you have trial aligned data that would be ideal for this group with timestamps of when ITIs, sound cues, and rewards happen. If not maybe information about reward will be sufficient for them for now (thanks for clarifying the rest of these questions). If there was an array of spikes and arrays of timestamps of different trial events they could also extract any information they are interested in themselves but I can see that organising the data would probably end up taking up a lot of their project time and potentially isn’t necessary for them. Thanks a lot for your help! :slight_smile:

@pachitarium would it be possible to add information on the average spike waveform (or some key features such as half-width, repolarisation time, etc…) for each unit?

@nsteinme is ‘response time’ the time of earliest detected movement after the go cue?

1 Like

@neurochong No sorry, there are no such recordings.
@veronika.samborska I will add the sound cues (i…e the go cue) and rewards/negative feedback today. ITI is already there (always 500ms), but I will make it more explicit.
@FranciscoSacadura I can add the waveforms and maybe compute half-width. Response time is indeed the earliest detected movement after the go cue, but keep in mind the go cue comes fairly late (at least 500ms after stim onset). There is no penalty for the mice turning the wheel before the go cue, so the earliest movements are almost always 200-300ms after stim onset, and almost always in the same direction as the detected “response” after the go cue.

1 Like

So the animal sees, they move, which generates movement in the stimulus, and the animal sees the movement. So activities in VIS is partly determined by behaviour. Is this right?

@kevinwli, You are right, but be careful because only movements after the go cue result in a movement of the stimulus. Before the go cue, the stimulus is locked in place, even though the mice are turning the wheels vigorously.


Ah, that helps. Thanks!

@pachitarium thanks! I was wondering then, if there are just pure behavior data that shows how fast or slow each mouse learned the task during the training?
a NMA project we were thinking of: to relate these possible individual differences in initial learning rates, to later recorded activity patterns during the actual task

That’s a good idea, but we don’t have any data during learning, and I doubt it would be easy for @nsteinme to dig that out.

There are however differences between mice with respect to reaction time distribution and types of errors. Could you use that instead?


yes that would work–thank you so much for the prompt replies!

Could we get data on which mouse ID corresponds to which trial? Also, how would we account for such a random variable in the data?

1 Like

hi @gail-harmata, do you mean which of the 39 sessions corresponds to which of the 10 mice? I think I omitted that, I should add it!

1 Like

yes, that’s what I meant, thanks!