Median value of channels subtracted to denoise in Neuropixels dataset

Hi, I read from the white paper that
"For each sample, the median value of channels N:24:384, where N = [1,2,3,…,24], was calculated, and this value was subtracted from the same set of channels. "
Is this why some negative peaks could be seen when ploting signals from certain channels at the beginning of stimulation and if these peaks could be recover using only the LFP data available through the AllenSDK?

Yes, the median subtraction procedure could explain a negative residual at the time of stimulation.

Can you post an image of what you’re observing? Median subtraction is only applied to the 30 kHz “AP band,” and the raw data for this is not available via the AllenSDK. But it can be accessed in other ways, for example as an AWS public dataset: Allen Brain Observatory - Visual Coding AWS Public Data Set - Registry of Open Data on AWS

Thanks and here is an example. Each line represents the signal of each channel within one second after drifting grating stimulation (average of all trials with contrast=0.8)

Hi @Jsl – since this is the LFP band, it’s not affected by the median subtraction procedure you mentioned. The LFP data is filtered above 0.1 Hz (to remove any DC offset) and then referenced to channels outside the brain. More details about the preprocessing methods can be found on this page: Visual Coding – Neuropixels — Allen SDK dev documentation

The activity you’ve plotted is the physiological response to the onset of a visual stimulus. It takes about 40 ms for stimulus-evoked spikes to reach VISal, which is the large negative peak seen at the beginning of the plot. The fact that some channels are positive-going and some are negative-going reflects the location of current sources and sinks across the cortical depth.

For a more detailed explanation of the mechanisms behind this, I’d recommend checking out this paper, which compares activity in a computational model to the LFP and spikes in thei Visual Coding dataset.