Computational Models in Neuroscience: How real are they? A Critical Review of Status and Suggestions

Review Article

Austin Neurol & Neurosci. 2016; 1(2): 1008.

Computational Models in Neuroscience: How real are they? A Critical Review of Status and Suggestions

Mohamad ASK¹ and Mohsin Reza²*

¹New Hearing Technologies Research Center, Baqiyatallah University of Medical Sciences and Department of Bioelectrics, Amirkabir University of Technology, Iran

²Section of Neuroscience, Department of Neurology, Baqiyatallah University of Medical Sciences, Iran

*Corresponding author: Mohsin Reza, Section of Neuroscience, Department of Neurology, Faculty of Medicine, Baqiyatallah University of Medical Sciences, Iran

Received: June 15, 2016; Accepted: July 29, 2016; Published: August 09, 2016


Computational models have become an important tool in the study of the nervous system and are commonly used in the simulation of specific aspects of physiology and pathology at various levels in addition to teaching neuroscience; however, their correspondence to complex biological reality of nervous system remains under question. The large number of variables and factors affecting the structure and function of the nervous system makes it almost impossible for them to be considered in a single model. Therefore, computational models, mostly consider few specific biological variables leading to their limited applicability. This paper provides some of the fundamental characteristics of the nervous system, including different cell types and their ratio, neurotransmitters and their receptors, synaptic plasticity and gene expressions to illustrate the gap between modeling studies and biological reality. Nevertheless, computational models have their own advantages that make them almost an irreplaceable tool in modern neuroscience. Specially, the integrative power of these models in unifying biological observations ranging from sub-cellular to the whole organism, have been attracting a lot of interest. Within this context, we critically review the biological correspondence of computational models and suggest multi-level modeling as an effective approach for enhancing the applicability of computational modeling in neuroscience.

Keywords: Computational Neuroscience; Neural Dynamics; Neurobiological Plausibility; Abstraction Level; Multilevel Modeling


Computer modeling of neuronal, glial, synaptic and network function has become increasingly popular during the last few decades and has greatly contributed to the understanding of various aspects of the nervous system in health and disease [1-3]. Models of neurons and glia exist that not only simulate selected output of neuronal function, but also predict cell function in different states. They also contribute in teaching various fields of neuroscience such as computational neuroscience, cognitive neuroscience, neural and cortical microcircuits, circuits and networks [4-6]. However, given the complexity of the nervous system, these models, especially the neuronal models have inherent deficiencies and are often criticized for their oversimplification and limited applicability [7-13].

This paper provides a critical overview and assessment of some of the inherent limitations of modeling studies of the nervous system from neurobiology perspective. After considering these limitations, a multi-level approach is suggested for bringing these models near to what exists in reality at cellular, receptor, genetic, molecular and network level. Additionally, for optimizing the effectiveness and application of these computational models, integration of experimental observations at different levels from cell to whole organism is suggested. As we will discuss, this across-scale models are not possible unless scientists from different fields and relevant disciplines including neurophysiologists and computational neuroscientists work in close collaboration. This will fill the gaps created by the lack of knowledge, diversity and can accurately connect the various levels of the hierarchy from single neuron to whole brain which is a common problem in models that are currently available

For this purpose, the remainder of the paper is divided into two main parts. First, some basic features of the nervous system such as cell types and their physiological aspects, synaptic plasticity and genetic considerations are discussed. Specifically, the gap between biological reality and modeling studies is explained; however, since this is not an exhaustive review, for each feature some examples of modeling studies pointing to their deficiency are provided for illustrative purposes. In the second part, from a modeling perspective, two main levels of modeling and their advantages and disadvantages are discussed and it is also explained how the multi-level approach may bring the advantages of both abstraction levels together in a hierarchical framework. Additionally, general requirements of any modeling study are suggested. The paper ends providing a general perspective based on the material presented in these two sections.

A Biological Perspective

There are several studies that have modeled neuronal function from various aspects and aimed to better predict the neuronal behavior under diverse conditions that would mimic a real neuron. However, while most of the investigators assert that their model is technically better than the others, these models have their own limitations as they consider some specific aspects of neuronal function [11]. To illustrate further, every modeling study usually focuses on one, two or few main variables and considers the effect of their variations on neuronal function as a change in the behavior of whole neuron or network of neurons and interprets relevant data as such. As we will discuss later, the possible number of variables that determine neuronal behavior under physiological and pathological conditions is too large and no single model can employ all these variables and their interactions. Such a model would require a huge computational effort as well as diversity of multi-disciplinary expertise to be constructed and analyzed. Therefore, from neuroscience view, the interpretation of the results of any such modeling study, especially when making general assessment, requires caution and has limited application.

In this section, some of the main physiological characteristics of the nervous system that are usually not addressed adequately in modeling studies, are investigated in order to point out the gap between complex biological reality and computational models and therefore the limited applicability of these models (Figure 1). To address this important problem, three specific considerations regarding the nervous system, namely, different cell types and their ratio, synapses and neurotransmitters and genes will be presented and few related modeling studies will be discussed.