restrict zeta to rate increases

[asumser]

Hi Jorrit!
I am currently implementing zeta in our analysis to have a robust way of determining cell responsiveness and the provided code works great! Thank you for putting this together.

In one of our specific use-cases we can however only consider neurons that have a significant rate increase with the stimulus. As I understand the code, vanilla zeta checks for deviations from baseline in any direction. Some neurons in our datasets are inhibited by the stimulus (based on the PSTH/IFR) with significant zeta but should be excluded. sRate.dblPeakTime is then nearly instantaneous for example. Do you have a hint how to best implement this? For now I am comparing sRate.dblPeakRate to baseline and check if sRate.dblOnset is close to sZETA.dblPeakT or sZETA.dblPeakT_InvSign. I am worried however that I am including neurons that have significant inhibition and a spurious excitatory peak. It would be great if you could offer an opinion, even if it is the case that zeta cannot be used here.

Thanks a lot and looking forward to the trace implementation for our Ca data!

[JorritMontijn]

Hi, thank you for your post. I think the problem boils down to how you would define "significant rate increase with the stimulus". To me, this could encompass anything ranging from Poisson-spiking neurons that show a rate increase during the stimulus (as we simulated in the paper) to neurons that show any significant, time-locked behavior where the # of spikes during the stimulus (say 1 second) is n+1, where n is the number of spikes during the baseline period of the same duration. So you'd have to draw a line somewhere defining what you find a relevant modulation. If this latter example is something you'd want to include, one option would be to calculate zeta, and for all significant cells then take the average rate during the stimulus, and only include them if the rate is higher than during the baseline. If you'd rather discard this "(n+1) vs n" cell as well, then where would you put the threshold? In any case, if there are specific functional subtypes you're looking for, perhaps making a scatter plot of the average rate difference between baseline and stimulus on one axis versus zeta-p values on the other might show you some clustering, which could give you an idea of where to put the threshold.
Either way, I would probably not use dblPeakT/dblPeakT_InvSign for this, because they only refer to the points in time at which the cumulative sum is statistically most anomalous, and are therefore difficult to interpret. For example, if you have a cell with a constant background firing rate and single onset peak of activity, both these points will lie close together and at opposite sides of the onset peak, with the onset time (and peak rate) in-between. In this case, whether dblPeakT or dblPeakT_InvSign is closer to your onset is not telling you anything (except perhaps whether the ascent is steeper than the descent).
I hope this helps, and let me know if you have any further questions!

[JorritMontijn]

This actually popped back in my mind again last night. Perhaps one definition of "rate increase" that could work here is whether the mean rate during the anomalous period that lies between dblPeakT and dblT_InvSign is higher than the background rate (for example, all other epochs that do not lie between +ZETA and -ZETA (or -ZETA and +ZETA, depending on their arbitrary signs)). You could still get statistically significant cells that show only a marginal rate difference, but perhaps this would help?

[asumser]

Hi, thank you for thinking on this until now. Good idea to use zeta to compare spiking in the anomalous period. Will try if it works on our data.