[NEW PROJECT]: Implement SpikeGLX gfix in spikeinterface

[alejoe91]

Project title: Implement SpikeGLX gfix in spikeinterface

Key Investigators

• Alessio Buccino

Project Description

SpikeGLX provides the gfix tool to automatically detect and blank out artifacts with large transients.

"-gfix=0.40,0.10,0.02. This requires that a transient artifact must attain a peak amplitude of at least ||0.40 mV||, must reach its peak as least as rapidly as ||0.1mV/sample-tick||, and the transient is over when the amplitude settles back to ||0.02 mV|| or smaller"

Background

See https://billkarsh.github.io/SpikeGLX/help/catgt_tshift/catgt_tshift/

Objectives

Port gfix to spikeinterface (probably as an alternative implementation of remove_artifacts or remove_transient_artifacts

Approach and Plan

• Implement GFIX in a lazy fashion
• Handle margins

Progress

Fill in with tentative milestones:

• Make PR to SpikeInterface
• Add documentation in the docs of SpikeGLX VS SpikeInterface preprocessing

[alejoe91]

@oliche want to help out with this project? :)

[harshk95]

Hi @alejoe91 ! Was wondering if there was any update on the implementation of gfix and/or the timeline for the same!

Thanks a lot!

[hkneu-95]

Has anyone managed to have an implementation of this?

[alejoe91]

Unfortunately no :( I think that @yger put together something similar though here: SpikeInterface/spikeinterface#3715

[oliche]

I have implemented this for the IBL pipeline, it is quite straightforward:
https://github.com/int-brain-lab/ibl-neuropixel/blob/2328394f08760bfec3d2042d1bc036c8cdc5fc48/src/ibldsp/voltage.py#L239

There are 2 important points for this to be effective:

• a cosine taper is important for any subsequent processing, even if just a few samples as in this implementation (7 by default)
• the big trick is to do this in 2 steps:
  • i) saturation detection and mute function computation on the raw raw data (even before DC offset application)
  • ii) only apply the mute at the end of the pre-processing (before whitening is ok but certainly after CAR)

Happy to help, but with the implementation above and this knowledge, it should be straightforward !

[alejoe91]

Nice! I have also implemented a saturation detector that finds positive/negative saturation events (knowing the saturation voltage).
https://github.com/AllenNeuralDynamics/aind-ephys-processing-qc/blob/main/code/qc_utils.py#L1321

@oliche @yger maybe we could integrate each others' approaches and finally implement something like gfix in spikeinterface?

[oliche]

Hi Alessio, I can give it a go !
I would start with the tests, could you point me to where you think I should put those ?

[alejoe91]

Hey @oliche

I would put both the function and tests in the spikeinterface.preprocessing module