Negative peaks dominate latency calculations

[jsiegle]

I would like to use the zetatest function to calculate the onset latency of a neuron's response to an optogenetic stimulus. However, the latency values that are being returned (e.g. dRate[dblLatencyPeak], dRate[dblLatencyPeakOnset]) do not make sense. Eventually I tracked this down to the fact that the negative peaks in the z-scored firing rate profile are often more prominent than the positive ones. This can be seen in the example below, where the original and inverted z-scored time series are plotted along with the peaks found by scipy.stats.find_peaks. The vertical blue line represents the most prominent peak, which is being detected as the first peak in the negative z-scored time series. Note that I restricted the temporal range between 0.002 and 0.015 s, which is where we expect the peak to occur.

I was able to correct this by ignoring the negative peaks in the getPeak function. After that, the returned latency values were more sensible.

Is my fix reasonable for cases where we know the firing rate should increase? Is there a way to apply the same constraints using the existing package?

[saman-abbaspoor]

Hi,

I'm using the Zeta test to find the latency and magnitude of peak responses. However, the values from vecLatencies and vecLatencyVals don’t seem to make sense (see attached). I could bypass the built-in peak detection in ZetaPy and perform it myself, but I wanted to ask whether doing so might cause any issues later on, and if it would be simpler to fix this within ZetaPy instead.

Thank you,

[heimel]

Hi, thanks for your question! For determining the onset latency of a response, we have now made a new function latenZy (http://github.com/Herseninstituut/latenZy). It is based on the zeta-test. By repeatedly applying a variant of the zeta-test at shortening intervals, if finds the last 'spontaneous' spike, and it works without extra parameters well on a range of data sets. The details are available in a preprint https://www.biorxiv.org/content/10.1101/2025.06.30.662308v1 .

The latency of the zeta-peak and negative zeta-peak don't mean much. The peak and half-height latency of the instantenous firing rate should give meaningfull values, but in your example they don't. Could you send me the example spike train and event times at my gmail address (alexander.heimel), so that I can determine what is going wrong and fix the bug? Regards, Alexander