Visual Data Mining of Brain Cells

Keywords: Hippocampal pyramidal cells, Electric stimulation, Morphometric parameters, XGobi

Abstract: Little is known about the influence of morphological variablity on the physiological response of brain cells. It is assumed that variables such as the number of branching points in a cell, its area and the change in shape from soma to terminals may influence the cell behavior differently under different electric stimulation.

It is difficult to separate the physiological differences in the cell (i.e. channel types and their distributions) from the morphological differences in laboratory experiments. Therefore, we have used computer simulations of 3-D neuroanatomical data of hippocampal cells while keeping the physiology across different cells constant. This allows us to look at the morphological influence on the neuronal function.

We have obtained data from neuroanatomical archives; specifically hippocampal pyramidal cells. Because of the amount of painstaking work to reconstruct the three dimemsional structure of a neuron, these archives tend to be very small (ranging from 3 - 50 cells). Additionally, there is the possibility that different archives have significant morphological differences due to the reconstruction technique. Therefore, it is difficult to reach robust conclusions on the morphological influence.

However, we found that a visual data mining approach with packages such as XGobi is a very effective way to detect some structure in the data. Morphometric parameters could predict both the firing rate and the firing behavior of our simulated neurons. There are, indeed, morphological variables that can be used to determine whether hippocampal pyramidal cells spike, burst, or have a plateau in response to a given level of input current.