search

MouseLand / suite2p

cell detection in calcium imaging recordings
http://www.suite2p.org
GNU General Public License v3.0
344 stars 240 forks source link

Python deconvolution implementation seems very different from OASIS and old Matlab implementation #996

Closed RandallJEllis closed 4 months ago

RandallJEllis commented 1 year ago

Sorry if I'm missing something obvious, but the deconvolution implementation appears to be quite different from the original OASIS implementation and the original C/MATLAB version of Suite2p. As far as I can tell, the Suite2p preprint and JNeuro paper are written about the MATLAB implementation, but I'm wondering if the newer, seemingly more minimalistic Python implementation is written about anywhere, specifically how it differs from the original OASIS method (aside from not inferring spike trains or imposing L1 penalties)?

Is the deconvolved output supposed to show:

  1. Underlying calcium activity after removing the influence of the sensor and frame rate?
  2. Underlying calcium transient amplitudes
  3. Underlying action potential amplitudes
  4. Something else?

Any clarification is greatly appreciated.

Hananel-Hazan commented 1 year ago

piling on @RandallJEllis question I open discussion post #1001

carsen-stringer commented 4 months ago

We re-implemented OASIS (this function here: https://github.com/j-friedrich/OASIS/blob/6dae97381b295f764530d952a3b2bf5033a009b1/oasis/oasis_methods.pyx#L25) in numba without L0 and L1 penalties. The OASIS paper discusses the interpretations of spike deconvolved traces, and we discuss this in our review paper, it is correlated to the firing rate of the neuron: https://www.sciencedirect.com/science/article/pii/S0959438818300977