Existing Methods to Reconstruct Neural Connections.
There
are several existing methods that are published. In this post I’m going to
describe some of them.
- Neural connectivity analyzing using Connectivity Matrix construction [1]
As
I said in the previous post, we gather the pairwise neural connections using
calcium florescence imaging [2]. And using this method we can discover and
analyze the neural connectivity. At the end of the post I will mention the
references and if you interested in this method you can refer them.
This method uses Matrix completion methods and local
thresholding to reconstruct neural connections. Basically this method has three
steps.
- Connectivity Matrix Semi-Construction
Suppose there are N neurons are in the sample, then we use NxN dimension
matrix to denote the pairwise connections that identified from the Calcium Florescence
Imaging. But in this step we enter only fraction of data to the connectivity matrix.
The completion of other half of the matrix will done in the next step.
- Connectivity Matrix Completion
To complete the rest of the matrix we use matrix completion techniques.
There is a fact that we need to consider, many times neurons are connected in
same way. As an example two different neurons may connect to the same neuron.
Using this fact and matrix completion techniques we can deduce an equation to
complete the rest of the matrix.
XM-calculated
entries in the connectivity matrix
ZM-corresponding
entries of the connectivity matrix approximation Z
||.||* Used to illustrate
nuclear form [3]
And we can use convex optimization solver to solve this equation
[4]. Note that solving this equation is much cheaper than calculating pairwise
connectivity scores. So that if we complete the other half of the matrix by
using this equation we reduce the half of our computation.
Local
Thresholding
- Local Thresholding
Using Local thresholding we can analyze the connectivity matrix
[5], [6]. We apply the following equation to each neuron we can identify the
connections of each this will help us to reconstruct the neural network.
2. Transfer
Entropy Method [7]
This method also gather data to analyze from Calcium Imaging.
This method basically have three steps.
- Time Series Chart
From the data that gather from the Calcium Imaging we can draw
the time series diagram for each neuron. The time series chart is the
illumination inside the neuron with respect to time. We use this charts to
analyze each neurons with other neurons.
- Transfer Entropy
We use following equation to calculate the transfer entropy. Consider two neurons as x and y.
Where,
Xt – Value of the time series of x at time t
Yt - Value of the time series of y at time t
This equation calculates the next value of sequence of x with respect to its own history and get the deference between the next values of x with respect to history of y. If x do not depend on y, then TE = 0 and otherwise TE>0. After some modifications we can construct the neural network.
References,
[1]Fast
algorithm for neural network reconstruction, Sean Bittner, Siheng Chen &
Jelena Kovacevic, Available; http://repository.cmu.edu/cgi/viewcontent.cgi?article=1309&context=ece
[2]Imaging
Calcium in Neurons, Christine Grienberger & Arthur Konnerth, Available;
http://www.sciencedirect.com/science/article/pii/S0896627312001729
http://www.sciencedirect.com/science/article/pii/S0896627312001729
[3] Nuclear
norm. G.L. Litvinov (originator), Encyclopedia of Mathematics. Available;
http://www.encyclopediaofmath.org/index.php?title=Nuclear_norm&oldid=19242
http://www.encyclopediaofmath.org/index.php?title=Nuclear_norm&oldid=19242
[4] E. J.
Cand`es and B. Recht, “Exact matrix completion via convex optimization,”
Journal Foundations of Computational Mathematics, vol. 9, no. 2, pp. 717–772,
Dec. 2009.
[5]
Normalized Iterative Hard Thresholding for Matrix Completion, Jared Tanner
& Ke Wei, Available;https://people.maths.ox.ac.uk/tanner/papers/TaWei_NIHT.pdf
[6] A Singular
Value Thresholding Algorithm for Matrix Completion, Jian-Feng Cai, Emmanuel J.
Cand`Es &Zuowei Shen, Available; https://statweb.stanford.edu/~candes/papers/SVT.pdf
[7] Reconstructing Neuronal Connectivity from Calcium
Imaging Data Using Generalized Transfer Entropy, Jina Li, Available; http://elischolar.library.yale.edu/cgi/viewcontent.cgiarticle=1178&context=ysphtdl
Comments
Post a Comment