Skip to content

Latest commit

 

History

History
71 lines (47 loc) · 3.41 KB

README.md

File metadata and controls

71 lines (47 loc) · 3.41 KB
Raw

Median Subtraction

Calls an executable that removes the DC offset and common-mode noise from a spike-band continuous file.

Because noise on Neuropixels probes is highly correlated across sites that share an ADC, we compute the median of every 24th channel, rather than using the median across all sites. This ends up creating a residual on the order of a few microvolts for large spikes, which can appear in the mean waveform. However, this is well below the probe's noise floor, and shouldn't affect spike sorting or data analysis.

SpikeInterface implementation

Performing median subtraction (and other preprocessing steps) is much easier with SpikeInterface. For Neuropixels data, we recommend using the "phase shift" to align the samples across channels, followed by median subtraction for the whole probe. This removes noise more effectively than just computing the median for simultaneously sampled channels.

Here is the code needed to preprocess Neuropixels data:

import spikeinterface.full as si

# read Open Ephys dataset
raw_rec = si.read_openephys('/path/to/data',
                            block_index=0,
                            stream_name='ProbeA-AP')

# or SpikeGLX, by changing one line of code:
raw_rec = si.read_spikeglx('/path/to/data', 
                           stream_name='imec0.ap',     
                           load_sync_channel=False)

# perform high-pass filtering (especially important for Neuropixels 2.0 data):
rec_filt = si.highpass_filter(raw_rec, 
                              freq_min=400.)

# remove channels outside the brain
bad_channel_ids, channel_labels = si.detect_bad_channels(rec_filt)
rec_cleaned = rec_filt.remove_channels(bad_channel_ids)

# perform the phase shift (similar to IBL destriping or `tshift` option in CatGT):
rec_shifted = si.phase_shift(rec_cleaned)

# subtract the median across all channels
rec = si.common_reference(rec_shifted, 
                          operator="median", 
                          reference="global")

The rec object can now be passed to the run_sorter() function to perform spike sorting.

More information can be found in the tutorial for analyzing Neuropixels data.

Dependencies

C++ source code for the median subtraction binary is available in the SpikeBandMedianSubtraction folder. This must be compiled prior to running this module.

Running

python -m ecephys_spike_sorting.modules.depth_estimation --input_json <path to input json> --output_json <path to output json>

Two arguments must be included:

  1. The location of an existing file in JSON format containing a list of paths and parameters.
  2. The location to write a file in JSON format containing information generated by the module while it was run.

See the _schemas.py file for detailed information about the contents of the input JSON.

Input data

  • AP band .dat or .bin file : int16 binary files written by Open Ephys, SpikeGLX, or the extract_from_npx module.
  • probe_info.json : file written by depth_estimation module.

Output data

  • AP band .dat or .bin file : overwrites the existing file with the median-subtracted data.
  • residuals.dat : contains the subtracted signals, which makes it possible to reconstruct the original data if necessary.