Modeling the nervous system

Abstract

he collection of articles published in this special issue of TINS is united by a common interest in the use of models as tools to unravel the complex workings of neural systems. As a biologist, I believe that the incorporation of such modeling techiques into the repertoire of basic neuroscientific research is absolutely essential for the eventual success of the field. Even in so-called 'simpler systems' it is clear that great amounts of detailed physiological and anatomical data alone are not enough to infer how these neural circuits work. Nervous systems are simply too complex to be understood without the quantitative approach that modeling provides. The papers in this issue were selected to demon- strate the considerable diversity of model-based approaches currently in use in neurobiology. The majority of these articles describe computer-based modeling efforts intent on replicating stimulus-induced neuronal responses or neurally controlled behavior. As such, they represent the modeling mainstream. However, the article by Bialek and Rieke (pp. 428-434) describes techniques used to decode the information content in single neuronal spike-trains - using models that learn to reconstruct the original stimulus from known neuronal responses. As the authors point out, this is the opposite objective of most modeling efforts. In the paper by Bulloch and Syed (pp. 422-427) techniques to construct what are essentially models of neural circuits using living cells in tissue culture are described. The argument is made that these reduced systems make it much easier to manipulate and understand the complex response properties of neur- ons, in addition to those of small networks. As mentioned above, the majority of the articles in this issue describe more conventional efforts to construct computer models of neuronal function. However, even in this category, the articles selected demonstrate a range of modeling approaches and assumptions. These fall into several distinct, though overlapping, categories that are also characteristic of modeling efforts in general. First, the articles represent the wide variation in the focus and scale of neurobiological modeling. At one end of this scale, the article by Kawato and Gomi (pp. 445-453) describes a systems-level approach to understanding the vestibular-ocular reflex. Their effort is strongly influenced by engineering theory and 'connectionist' modeling techniques, and is intended to explain the behaviorally related interactions of vast numbers of neurons found in several different mammalian brain structures. At the other end of the continuum, Segev (pp. 414-421) describes the tech- niques used to simulate single neurons in great anatomical and physiological detail. Between these two extremes lies the article by Calabrese and De Schutter (pp. 439-445), which describes a bio- physically realistic model of several interacting in- vertebrate neurons, and the contribution by Cohen and colleagues (pp. 434--438), which describes an ongoing multi-lab collaboration intent on under- standing the behavior generated by many more neurons within the lamprey spinal chord. A second difference displayed by the models presented involves their degrees of simulated bio- logical realism. The central pattern generator (CPG) model described by Calabrese and De Schutter includes considerable biological realism at the level of single neurons. In contrast, the CPG modeling of Cohen et al. is based on a mathematical simplification that effectively removes the biophysical behavior of single cells from the model. In both cases the models make biologically testable predictions and provide biologically plausible interpretations. Somewhere in the middle, the article by Segev starts with the assumption that biophysically detailed models of single cells are necessary, but that these models should be used to generate more reduced representations of single cells for eventual inclusion in network simulations. Finally, these modeling efforts, like those in the wider literature, can be very broadly grouped into those that are intended to demonstrate the biological plausibility of a particular pre-existing idea and those that are used as tools for obtaining and testing new ideas. For example, in the article by Kawato and Gomi, the modeling efforts described are strongly influenced by more abstract considerations of motor performance as well as by work with more abstract network models. In this case, the modelers' task is to implement abstract ideas in as biologically realistic a manner as possible. In contrast, the article by Calabrese and De Schutter demonstrates how new ideas about the organization of neural systems can emerge from the process of constructing a biologically realistic model. In this case, the model becomes just another tool the physiologist uses to explore the properties of a complex and not yet well understood circuit.