# Connectomics – Inferring Connectivity From Calcium Fluorescence

What's the

**connectivities**in population of neurons ?`connectomics`

What's the

**mathematical model**of spikes ?`mathematics`

How do neurons

**comupte**and**memory**?`intelligence`

We are working towards these exciting problems !

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

`Fast-oopsi`

was developed by Dr. **joshua vogelstein** in 2009, which is now widely used to extract neuron spikes from calcium fluorescence signals. Here, we propose detailed implementation of the fast-oopsi algorithm in the python programming language, which is called `py-oopsi`

. Some corrections are also made to the original fast-oopsi paper.

* The py-oopsi implementation can be obtained at: py-oopsi @ github *

**Top (a),** The synthetic calcium fluorescence trace, random noise are added to the ground truth signal (shows in red color). The parameters used to generate the synthetic fluorescence signals are : T=2000, dt=0.02ms, lambda=0.1, tau=1.5 and sigma=0.2.

**Bottom (b), (c), (d),** The reconstructed spikes from fluorescence trace using `py-oopsi`

, `wiener filter`

, and `discretize binning`

. The py-oopsi favors sparse (by adopting Poison prior and Gaussian approximation) and non-negativity, it has the most sparse yet positive spikes. Wiener filter has oscillary at the inferred spikes. It can also reconstruct spikes correctly. We may, for example, set a threshold at 0.5 to obtain clean spikes. Discretize with binning is not robust agains noise. However, in stetter2012, he used discretized with binning and GTE (generalized transfer entropy) to achieve ~0.90 AUC.

## Connectomics challenge at kaggle

This connectomics challenge brought about many interesting discussions, it also draw me a new look (perspective) on the frontier of the neuron technology. For example the Netword Deconvolotion (ND) method, and many usefull features used by the kaggler. This section will address more of the related issues on the forum.