(A) simulation study on the modulation of information transfer in feedforward networks

Abstract

There are vast amounts of correlated neural activities, such as oscillations and synchronizations, observed in the brain. These activities are one of the most important keys for communication in the brain. However, the questions on how the synchronization level can be modulated and how these synchronized activities affect spike transfer from one layer to another layer in different convergent connection conditions are not clearly understood. In this work, we employ computer simulation of realistic neural network to address the questions.

The neural network consists of two layers of conductance based single cell model, the source layer and the target layer. The interlayer connections follow statistical wiring rule in which, strength and connectivity of the connection depend on the distance between the target cell and the source cells defined as the Gaussian-Gaussian (GG) convergent rule. For comparison, two other convergent rules that have constant probability of connection within the range were made. One rule has uniform distribution of connection strength. This rule is defined as the Uniform-Uniform (UU) convergent rule. Another rule has random connection strength that draws from negative exponential distribution and it is named as the Uniform-Exponential (UE) convergent rule. Then, the responses on target layer were measured from various convergent conditions. To study how the synchronization of input contributes to the neural network's response, this work compared the oscillating input and the static input given to the system.

This work observed that, when oscillating input were given, output firing rate is higher than that of static input even though the overall firing rate of both inputs were the same. Also, the specific level of response can be made with less strength of connection in oscillating input compared to the static input. This work showed that the oscillating input induced more output spike than static input. In addition, the gain of response from static to oscillating input is the largest in UU rule compared to GG and UE rules. We discovered that the output spikes in UU rule have the most dependency on synchronization level in input compared to GG and UE rules. These results showed that UU has the most sensitivity on the synchronization level in input.

In summary, this work found that high level of synchronization in input results in high output response and the synchronized input can make a specific level of output with low cost of connection compared to the static input. Also, neural network with Uniform-Uniform convergent connection rule is selective to change in input synchronization compared to the other two rules. These results suggest that both input synchronization level and the interlayer connection rules contribute to understanding brain connections.