Computational Neuroscience

from Wikipedia, the free encyclopedia

Computational neuroscience (from computation : calculation, information processing, and neuroscience : neuroscience, brain research; mostly synonymous with theoretical neuroscience ) is an interdisciplinary science that deals with the information- processing properties of the nervous system . Information processing means the entire spectrum of brain functions from the various stages of processing sensory impressions to cognitive functions such as learning , memory , decision-making and the control of the motor system for executing actions .

The most important methodological tool in Computational Neuroscience is the mathematical modeling of components of the nervous system such as nerve cells , synapses and neural networks using the methods and knowledge of biophysics and the theory of dynamic and complex systems . Due to their complexity, these models are often simulated in the computer . In addition, Computational Neuroscience also provides methods of analyzing experimental neuronal data. All these approaches require close collaboration between experimental scientists from the disciplines of biology , medicine , psychology and physics as well as theorists from mathematics , physics and computer science . The experimental data provide both the basis for the models (e.g. electrophysiological properties of nerve cells and synapses, network structures in real nerve networks) and the possibility of testing their predictions, for example about certain dynamic or information-processing properties. The models, in turn, offer the possibility of using the often diverse and B. to systematically organize apparently contradicting results of the experiments and to recognize complex relationships through mathematical analysis and simulation, which are difficult or impossible to grasp without this method.

The object of the modeling is structures on all scales of size and complexity, starting with biophysical simulations of the molecular dynamics of certain ion channels and neurotransmitters , through models of individual nerve cells, to complex network models that simulate interactions between brain regions . Depending on the question, these models can have very different degrees of abstraction . that is, either closely related to experimental data or rather depict and formalize the general principles and structures obtained from the experiments.

Computational neuroscience can to a certain extent be distinguished from connectionist theories of psychology, pure learning theories such as machine learning and artificial neural networks, and the field of neuroinformatics , although some of these areas have parallel development histories and some also pursue similar goals. Modeling approaches in computational neuroscience aim to map certain aspects of neuronal structures in a biologically realistic manner and to make direct predictions through corresponding experiments. Connectionist models pursue a similar prediction goal at the level of psychophysical experiments, but have only a limited claim to biological realism, which is limited to the structure of the connections and the ability to learn. The same applies to learning theories, which are often also used for purely technical purposes, for example for predicting a complex time series or for recognizing patterns in images. In these application-oriented areas, the analogy to the brain only plays a subordinate role; an understanding of human information processing is not sought. Finally, neuroinformatics, following its name, takes a computer science perspective on the neurosciences. This includes the development of databases , data structures and standards for efficient storage, archiving and exchange of experimental data as well as the development of software both for modeling neural systems (e.g. Neuron , Genesis , NEST ) and for recording and analyzing experimental data . More abstract approaches such as artificial neural networks and machine learning are sometimes also included in neuroinformatics.

Research topics

Electrical schematic for the Hodgkin-Huxley model

An early neuron model (1952) which, partially modified, is often the basis of today's software , is the Hodgkin-Huxley model . Based on a description of the electrical properties of the cell membrane of neurons, which are decisively influenced by ion channels, in the form of an equivalent circuit diagram , it models the emergence of action potentials . The mathematical methods used in the various models originate mainly from the theory of dynamic systems. The sometimes erratic behavior of neurons (e.g. in the area of ​​the threshold potential ) is taken into account by bifurcations .

Examples of the application of such models are the description of cells in the basal ganglia , with the aim of developing new therapeutic approaches for Parkinson's disease , in which the modeling of individual cells (e.g. possible with the Neuron software) is important and attempts to describe complex cognitive processes as in the Stroop test with the Emergent program , whereby additional effects such as Hebb's learning rule play a role, but individual cells are simplified significantly more because of the number taken into account.


The term “Computational Neuroscience” was introduced in 1985 by Eric L. Schwartz . At the request of the Systems Development Foundation, Schwartz organized a conference in Carmel, California that year. The aim of this was to provide an overview of a field of science that had previously been associated with a number of different terms such as “neural modeling”, “brain theory” or “neural networks”. The contributions to this conference were published in 1990 in a book called "Computational Neuroscience".

The early history of this area is closely linked to the names of scientists such as Louis Lapicque (1866–1952), Alan Lloyd Hodgkin and Andrew Fielding Huxley , Wilfrid Rall , David H. Hubel and Torsten N. Wiesel , and David Marr .

Lapicque introduced the Integrate-and-Fire neuron model in 1907, which is still one of the most popular models in computational neuroscience due to its simplicity. Almost 50 years later, Hodgkin and Huxley studied the giant axon of the octopus, which was particularly accessible experimentally, and derived the first biophysical model of the action potential (Hodgkin-Huxley model), which they published in 1952. Rall extended this model to include the cable theory , which laid the basis for neuron models that are composed of spatially extended parts of the cell ( soma , axon , dendrites ). Today such models are used for morphologically exact simulation, e.g. B. used with the help of Neuron .

Hubel and Wiesel researched the cells of the primary visual cortex , the first area of ​​the cerebral cortex that receives visual information from the retina . Among other things, they discovered that the cells of the primary visual cortex not only reflect the spatial structure of the image on the retina, but can also read out the spatial orientation of the perceived objects. Both Hodgkin and Huxley and Hubel and Wiesel received the Nobel Prize in Physiology or Medicine (1963 and 1981) for their work .

Marr's work focused on the interactions between neurons in different areas such as B. the hippocampus and the cerebral cortex . He presented a theory of vision based on the principles of electronic data processing in computers. He is considered to be one of the founders of neuroinformatics.

See also


  • Larry F. Abbott, Peter Dayan: Theoretical neuroscience: computational and mathematical modeling of neural systems . MIT Press, Cambridge, Mass 2001, ISBN 0-262-04199-5 .
  • William Bialek, Fred Rieke, David Warland, Rob de Ruyter van Steveninck: Spikes: exploring the neural code . MIT Press, Cambridge, Mass 1999, ISBN 0-262-68108-0 .
  • Alla Borisyuk, G. Bard Ermentrout, Avner Friedman, David Terman: Tutorials in Mathematical Biosciences 1: Mathematical Neuroscience: v. 1 . Springer, Berlin, Berlin 2005, ISBN 978-3-540-23858-4 .

Web links

Individual evidence

  1. ^ What is computational neuroscience? Patricia S. Churchland, Christof Koch, Terrence J. Sejnowski. in Computational Neuroscience pp. 46-55. Edited by Eric L. Schwartz. 1993. MIT Press Archive Link ( Memento of the original from June 4, 2011 in the Internet Archive ) Info: The archive link was inserted automatically and has not yet been checked. Please check the original and archive link according to the instructions and then remove this notice. @1@ 2Template: Webachiv / IABot /
  4. Alla Borisyuk, G. Bard Ermentrout, Avner Friedman, David Terman: Tutorials in Mathematical Biosciences 1: Mathematical Neuroscience: v. 1 (Lecture Notes in Mathematics) . Springer, Berlin, Berlin 2005, ISBN 978-3-540-23858-4 .
  5. ^ JE Rubin, D. Terman, High Frequency Stimulation of the Subthalamic Nucleus Eliminates Pathological Thalamic Rhythmicity in a Computational Model. In: Journal of Computational Neuroscience 16, 2004, 211-235
  6. stove, SA, Banich, MT & O'Reilly, RC (2006). Neural Mechanisms of Cognitive Control: An integrative Model of Stroop Task Performance and fMRI data. Journal of Cognitive Neuroscience, 18, 22-32.