this is a working document

Py-oopsi: the python implementation of the fast-oopsi algorithm

Fast-oopsi was developed by joshua vogelstein in 2009, which is now widely used to extract neuron spike activities from calcium fluorescence signals. Here, we propose detailed implementation of the fast-oopsi algorithm in python programming language.

Py-oopsi requires numpy, scipy and matplotlib.

Generate Synthetic Calcium Trace

To generate synthetic calcium trace, you can

T = 2000
dt = 0.020
lam = 0.1
tau = 1.5
sigma = 0.2

# signal generator
F,C,N = oopsi.fcn_generate(T, dt=dt, lam=lam, tau=tau, sigma=sigma)

where F is the Fluorescence signal with noise, C is the clean calcium trace, N is the ground truth spikes.

Reconstruct Spikes via py-oopsi

We provide demo.py to illustrate the usage of py-oopsi (as well as wiener filter, discretized binning),

# fast-oopsi,
d,Cz = oopsi.fast(F,dt=dt,iter_max=6)

# wiener filter,
d,Cw = oopsi.wiener(F,dt=dt,iter_max=100)

# descritized binning,
d,v = oopsi.discretize(F,bins=[0.75])

Tweak py-oopsi

py-oopsi requires

• F the fluorescence signal, a numpy.ndarray object of 1-D vector;
• dt the frame interval, 1/(frame rate);
• iter_max maximum number of iteration;
• update true if the parameters are updated after each iteration.

when imaging large population of fluorescene signals of multiple neurons, for example, the connectomics challenge at kaggle.com, you need to write a subroutine to process the fluorescence trace per neuron.

Reference

• Joshua T Vogelstein, Adam M Packer, Tim A Machado, Tanya Sippy, Baktash Babadi, Rafael Yuste, Liam Paninski [Fast non-negative deconvolution for spike train inference from population calcium imaging] (http://stat.columbia.edu/~liam/research/pubs/vogelstein-fast.pdf) Journal of Neurophysiology, 104(6): 3691-3704