Dataset: recording with a Neuropixels Phase3 electrode array

Recording, data processing, and documentation by Nick Steinmetz, nick[dot]steinmetz@gmail, in the CortexLab at UCL.

Many thanks to: Marius Pachitariu, Cyrille Rossant, Max Hunter, Chris Burgess, Matteo Carandini, and Kenneth Harris.

Please feel free to email me - N. Steinmetz - with any questions or comments.

About this dataset

Files are located here.

A "Phase3" Neuropixels electrode array was inserted into the brain of an awake, head-fixed mouse for about an hour. This electrode array recorded 384 channels of neural data at 30kHz and <7ÁV RMS noise levels; see Lopez et al. 2016 for further technical details. The sites are densely spaced in a "continuous tetrode"-like arrangement, and a whole array records from a 3.8mm span of the brain. The array recorded from visual cortex, hippocampus, and some parts of thalamus. After applying ocmmon average referencing, the data were spike-sorted automatically by Kilosort (software by M. Pachitariu; using included "master_file.m") and manually by N. Steinmetz using Phy (manual phase; software by C. Rossant). In total 242 well-isolated individual neurons were identified.

Example raw data:

Example neuron waveform and autocorrelogram:

In addition to neural data, a movie of the eye of the mouse was acquired at 100Hz and a movie of the frontal aspect (face, forearms, anterior trunk) was recorded at 40Hz. The mouse's forepaws rested on a wheel that could move left or right, and the movements of this wheel were recorded. Auditory and olfactory information was not recorded and the mouse was not guaranteed to be in silence or in an odor-free environment. Behavioral data available upon request.

Data was high-pass filtered at 300Hz on-probe. Low-pass, LFP data files also available on request.

The experiments that were run

During the first phase of the recording, the mouse was shown visual stimuli (described below) on three monitors placed around the mouse at right angles to each other, covering about +/-135 degrees azimuth and +/-35 degrees elevation. During the second phase of the recording, the monitors were turned off and the mouse was in complete or nearly-complete darkness.

Experiment 1: Sparse noise stimuli for receptive field mapping

8 degree squares were displayed at random times for 1/6 of a second on each vertex of a 9x34 grid with 8 degree spacing covering the whole extent of the monitors. Background was grey and squares could be either white or black. The presentation consisted of three repeats of a single five-minute long pattern of squares.

Experiment 2: Darkness, no stimulation

Loading the data

The raw data (".bin" files) are binary files of type int16 and with 385 data rows (see In matlab this can be read as follows:

>> fn = 'rawDataSample.bin';
>> fid = fopen(fn, 'r');
>> dat = fread(fid, [385 Inf], '*int16');
>> fclose(fid);
>> chanMap = readNPY('channel_map.npy');
>> dat = dat(chanMap+1,:);
>> figure; imagesc(dat(:,1:30000))

The preprocessed data is stored in a set of npy files. These can be loaded with numpy in Python, or you will need the npy-matlab repository to load them in Matlab.

You can view the data most easily with Phy, please see specifically this documentation for details about the meaning of the included files.


You may use this dataset for any research or personal purposes but please do not distribute or publish without my permission.


C. M. Lopez et al.,"A 966-Electrode Neural Probe with 384 Configurable Channels in 0.13Ám SOI CMOS," Solid-State Circuits Conference Digest of Technical Papers (ISSCC), 2016 IEEE International, San Francisco, CA, Paper#22.7, pp21-23