Trends in European Computational Neuroscience
C. & O. Vogt Brain Research Institute and Institute of Morphological Endocrinology and Histochemistry,
Heinrich Heine University, PO Box 101007, D-40001 Düsseldorf, Germany
European computational neuroscience
Structure-function relationships, neuroinformatics, computational neuroscience, neural
information processing, theoretical neurobiology, European initiatives, Forum of European
Address for correspondence:
Dr. Rolf Kötter
C. und O. Vogt-Institut für Hirnforschung
D-40225 Düsseldorf, Germany
Understanding the complex structure and function of the central nervous system
requires the integration of many scientific disciplines and descriptive levels, and
the development and application of adequate research tools. The fast growing
field of computational neuroscience plays an important role within this integrative
effort. In Europe, an active community is emerging that investigates structure-
function relationships in the nervous system at subcellular, cellular, network, and
behavioural levels. An increasing number of courses and conferences are
offering opportunities for development and exchange of ideas between scientists
interested in computational neuroscience. Thus, it is timely to present a spectrum
of European activities in this field and to foster the communication between
scientists across disciplines towards joint efforts.
Towards the end of the twentieth century the field of neuroscience has seen an
unprecedented expansion. Vast amounts of data and many noticeable insights
have been produced but we are still long ways from understanding how the brain
works. What becomes clearer is that this understanding is not attained in a single
step: like many other complex questions it has many different facets that need to
be addressed in detail. In analogy to the 23 mathematical problems presented by
David Hilbert in 1900 one can conceive a list of important and difficult, but
accessible problems that neuroscientists should strive to solve
(http://www.hirn.uni-duesseldorf.de/~rk/hilbert.htm). Although one may argue
about the relative importance of these various problems, decisive advancements
on any item on this list would constitute a major step in the right direction.
A common feature of these problems is that they address relationships between
morphological, physiological and behavioural aspects of information processing
in nervous systems ("structure-function relationships") as, for example, the
synaptic mechanisms of information storage or the role of anterior cingulate
gyrus in conscious decisions. This broad scope is a challenge since it requires
interdisciplinary approaches across established neuroscientific disciplines, such
as neuroanatomy, neurophysiology, neuropsychology, and clinical neuroscience.
Besides horizontal links between disciplines most approaches require vertical
cross-linking of different levels of description including molecular, cellular,
network or organismic levels. Since each level has its own set of theories and
methods it constitutes another challenge for neuroscientists to integrate these.
An increasing number of research publications in highly ranked neuroscientific
journals show aspects of such horizontal and vertical cross-linking and underline
the significance of integrated approaches to address important neuroscientific
questions. It is important to keep these questions in mind and not to lose oneself
when diving into the myriad details that need to be worked out before decisive
answers are produced.
How does integration across disciplines and across descriptive levels in
neuroscience take place? First of all, neuroscientists themselves make this
happen: anatomists have started to apply molecular approaches,
electrophysiologists embrace cell staining and network reconstruction,
neuropsychologists use pharmacological manipulations in functional imaging,
and clinicians venture into a broad range of basic neuroscientific disciplines.
Thus, the reality of research work carried out in the traditional disciplines is
changing towards a broader spectrum of methods and co-operation. This
tendency is reflected in an increasing proportion of funding and education
available for co-operative projects in neuroscience. In addition, an increasing
impact is made by scientists who have specialised into the investigation of
aspects common to several neuroscientific disciplines and levels of description,
such as principles of information processing and properties of non-linear
systems. Their backgrounds range even further, extending to engineering,
computer science or theoretical physics, and their levels of approach may be
even more abstract, employing information theory or phase plane analysis to
neuronal and brain functions. Eventually, the complexity of the brain seems to be
reflected in the complexity of expertise and methods employed to unravel its
These encouraging developments bear their problems, too. While some
departments or disciplines may fear to lose their identities or impact, others may
re-duplicate techniques that are available at higher standards elsewhere.
Different educational backgrounds may not only hamper interdisciplinary
communication, but also lead to biased opinions about what constitutes an
important neuroscientific problem or a valuable contribution towards its solution.
In particular, it seems counterproductive to argue about the superior significance
or authenticity of empirical versus theoretical approaches when both can provide
important insights or even complement each other. Therefore, we need to
develop a strong integrative neuroscientific culture by fostering exchange of
ideas, co-operation between research groups, interdisciplinary education of
students, and administrative integration of departments and disciplines.
Computational neuroscience is a representative of such an integrative
neuroscientific culture since it is located at the crossroads of all neuroscientific
disciplines across all levels of description combining empirical and theoretical
approaches. As the contributing disciplines and their subjects of study develop, it
is constantly changing itself requiring adjustment of subjects, concepts and
techniques. Thus, computational neuroscience is more adequately understood as
a mental attitude than a discipline. It may be instantiated administratively by
research groups, departments, collaborative research centres or priority research
programmes. Computational neuroscience necessarily has a broad scope and is
open to interpretations. Attempts to define it have failed in the past: it is neither
limited to understanding the computational aspects of neural processes, nor the
application of computers to the study of the nervous system, nor the combination
of the two, although both approaches are commonly mentioned and widely used
/8/. For the same reason it is useless to draw sharp distinctions between
computational neuroscience and similar concepts such as neuroinformatics,
theoretical biology, analysis of neural systems or neural information processing.
Such attempts serve mostly to emphasise a priority of certain contributing
disciplines, to focus on arbitrary pet topics or to appeal to some supporting
The focus of this Special Issue on computational neuroscience in Europe is such
an artificial and arbitrary limitation. This limitation does in no way intend to claim
a European priority or to depreciate contributions from other parts of the world.
On the contrary, it responds to a conspicuous lagging behind of computational
neuroscience in Europe compared to the United States where annual
computational neuroscience conferences
(http://www.bbb.caltech.edu/cns/cns.html) have seen an enormous growth since
their modest start in 1992. In response to the growing attendance of European
research workers it will be a big step forward for this meeting to take place for the
first time in Europe in Bruges in the year 2000. Similarly, the recent organisation
of EU advanced courses in computational neuroscience (http://bbf-
www.uia.ac.be/courses/CRETE/Crete_index.html; /10/) follows a scheme that
had been successful at the Cold Spring Harbor and Woods Hole Marine
Biological Laboratories for many years. What is missing in Europe, however, is a
visible community that actively advances the field through methodological
workshops, integrated teaching and, most importantly, through intense
communication and exchanges with other (in particular: experimental)
In connection with the Forum of European Neurosciences in Berlin 1998, which
successfully displayed a fresh spirit in European neuroscience /9/, two important
computational events took place. A satellite "Workshop on Neuroinformatics" was
organised (http://itb.biologie.hu-berlin.de/neurinf98.html) that facilitated
communication between computational neuroscientists and formation of plans for
European initiatives within the 5th
framework programme 1998-2002 under point
9.2 of the work programme "brain theories, computational neuroscience and
neuroinformatics". Such initiatives will hopefully allow for a continuation of the EU
advanced courses and for the formation of a network of research groups that are
committed to the advancement of methods, teaching and research in
The other event was an extended Special Workshop entitled "Spectrum of
Computational Neuroscience in Europe" (http://www.hirn.uni-
duesseldorf.de/~rk/ena98sw5.htm) as part of the main Forum of European
Neuroscience conference programme. The aims of this Special Workshop were
to present to a general neuroscience audience the efforts made by an active and
growing European community that investigates structure-function relationships of
the nervous system using computational neuroscience approaches, and to foster
the exchange of ideas between neuroscientists in different disciplines. The
immediate outcome of the meeting and subsequent discussions was the intention
to form a thematic network that would enhance the visibility of computational
neuroscience in Europe and distribute information via the world wide web to all
interested individuals and organisations. Further suggestions included the
formation of sections on computational neuroscience within the national
neuroscience associations and their European federation (FENS), and the
organisation of regular meetings in association with the Forum of European
neuroscience. This initiative was disseminated through computational mailing
lists (e.g. Comp-Neuro: firstname.lastname@example.org) and led to the creation of the
first comprehensive and informative web page on computational neuroscience in
In addition, the contributors to the Special Workshop were invited to prepare
peer-reviewed full articles towards a Special Issue of the journal Reviews in the
Neurosciences that would communicate current developments in European
computational neuroscience. As a result the present issue covers a wide range of
work carried out in several European countries ranging from subcellular to global
system levels and applying to a wide range of species.
Péter Adorján, Christian Piepenbrock and Klaus Obermayer in Berlin contribute a
model of “Contrast adaptation and infomax in visual cortical neurons” to explain
how the visual system can adapt to different levels of stimulus contrast /1/. The
proposed mechanisms constitute a bridge between abstract principles and basic
physiological findings on the function of synapses and on the phenomenon of
pattern adaptation. This work provides an important advancement after 20 years
of research on pattern adaptation in the visual system.
“Single cell and population activities in cortical-like systems” are the topic of
Fülöp Bazsó, Ádám Kepecs, Máté Lengyel, Szabolcs Payrits, Krisztina
Szalisznyó, László Zalányi and Péter Érdi in Budapest /2/. These authors raise
the question of the appropriate levels of description for gaining some
understanding of certain parts of the brain. They investigate aspects of neural
information processing in the hippocampus and olfactory bulb using three
differently detailed modelling approaches.
From Plymouth comes the work of Roman Borisyuk, Michael Denham, Susan
Denham and Frank Hoppensteadt on “Computational models of predictive and
memory-related functions of the hippocampus” /3/. The first model presents a
temporally asymmetric learning rule for the synaptic weights of the dentate gyrus
- CA3 projection that fits empirical data of CA3 pyramidal cell activity during
prediction of sensory events. The second model investigates memorisation of
sequences of sensory events as an effect of entorhinal and septal oscillatory
inputs on a chain of interacting neural oscillators in CA3.
Erik De Schutter at Antwerp has been “Using realistic models to study synaptic
integration in cerebellar Purkinje cells” /4/. Illustrated by examples from this work
he proposes a basic similarity of experimental and computational studies
("experiments in computo") which makes it possible to turn around the common
approach of using computer models for hypothesis testing, and to build realistic
neuronal models for generating testable hypotheses with considerable predictive
Rolf Kötter and Dirk Schirok at Düsseldorf review advances “Towards an
integration of biochemical and biophysical models of neuronal information
processing: a case study in the nigro-striatal system” /5/. They note a surprising
lack of subcellular models of information processing and demonstrate their
technical feasibility and scientific promises. Extensions to combined biochemical
and electrophysiological studies are discussed from the viewpoint of
experimental and theoretical work on the role of dopamine- and calcium-
mediated signals in the striatum.
“Interaction of cortex and hippocampus in a model of amnesia and semantic
dementia” is the topic of Jaap Murre from Amsterdam /6/. In an attempt to relate
psychological functions to underlying brain mechanisms he describes a highly
abstract network model inspired by certain features of these brain regions.
Lesioning different parts of the network causes typical alterations of its dynamic
behaviour that resemble symptoms observed in some types of memory
disorders. Inclusion of further details may help to validate the assumptions of the
model and to extend its predictive value.
A contribution from Trieste by Giulietta Pinato, Pietro Parodi, Alessandro Bisso,
Domenico Macrì, Akio Kawana, Yasuhiko Jimbo and Vincent Torre investigates
“Properties of the evoked spatio-temporal electrical activity in neuronal
assemblies” /7/. This study raises the important question how reproducible
information processing can be achieved in networks whose elements transmit
information with low reliability. Based on analyses of stimulation-evoked
responses in the leech ganglion and in neuronal cultures the authors conclude
that averaging or distributing activity across many neurones can be used to
decode the transmitted information faithfully in the presence of noise.
“What can robots tell us about brains? A synthetic approach towards the study of
learning and problem solving” is proposed by Thomas Voegtlin and Paul
Verschure at Zürich /11/. The problem of indeterminacy that modellers have to
face when attempting to validate integrative models has to be tackled by
imposing additional constraints from multiple levels of description. The authors
give examples from their work, which show that the implementation of robots
provides rich behaviours that are capable of constraining such models from an
additional macroscopic level.
Besides demonstrating the wide range and common features of computational
neuroscience the contributions to this issue identify areas that require further
development. These include the theoretical and practical advancement of
computational neuroscience particularly in the directions of modelling subcellular
biochemical and molecular processes, high-level phenomena as revealed by field
potential recordings and functional imaging, or ontogenetic and phylogenetic
developments. A culture and standards of computational modelling need to be
developed that include justification of the chosen modelling approach, control of
results by application of sophisticated statistical techniques and validation of
generalisations by exploration of parameter space. Interactions between
scientists who apply theoretical and experimental approaches need to be
intensified to improve the quality and impact of research work. Multi-modal
databases of experimental data and models need to be co-ordinated to facilitate
cross-fertilisation and to enable integrative work towards a better understanding
of how the brain works.
I wish to thank the organisers, advisors, and sponsors of the Forum of European
Neuroscience 1998 and the representatives of Reviews in the Neurosciences for
their substantial support.
1. Adorján P, Piepenbrock C, Obermayer K. Contrast adaptation and infomax in
visual cortical neurons. Rev Neurosci 1999; this issue.
2. Bazsó F, Kepecs A, Lengyel M, Payrits S, Szalisznyó K, Zalányi L, Érdi P.
Single cell and population activities in cortical-like systems. Rev Neurosci
1999; this issue.
3. Borisyuk R, Denham M, Denham S, Hoppensteadt F. Computational models
of predictive and memory-related functions of the hippocampus. Rev Neurosci
1999; this issue.
4. De Schutter E. Using realistic models to study synaptic integration in
cerebellar Purkinje cells. Rev Neurosci 1999; this issue.
5. Kötter R, Schirok D. Towards an integration of biochemical and biophysical
models of neuronal information processing: a case study in the nigro-striatal
system. Rev Neurosci 1999; this issue.
6. Murre JMJ. Interaction of cortex and hippocampus in a model of amnesia and
semantic dementia. Rev Neurosci 1999; this issue.
7. Pinato G, Parodi P, Bisso A, Macrì D, Kawana A, Jimbo Y, Torre V.
Properties of the evoked spatio-temporal electrical activity in neuronal
assemblies. Rev Neurosci 1999; this issue.
8. Sejnowski TJ, Poggio TA. Series Foreword. In: Koch C, Segev I, eds.
Methods in Neuronal Modeling. Cambridge: MIT, 1998; ix.
9. Singer W. Towards a European forum for the neurosciences. Trends
Neurosci 1997; 20: 116-118.
10.Ullman S, Roth A, Thomson A, Linne ML. Crete, channels, cells, circuits and
computers. Trends Neurosci 1997; 20: 53-54.
11.Voegtlin T, Verschure PFMJ. What can robots tell us about brains? A
synthetic approach towards the study of learning and problem solving. Rev
Neurosci 1999; this issue.