Abstracts

Journal articles and book chapters

Frégnac Y, Rudolph M, Davison A.P. and Destexhe A. (2007) Complexity in Neuronal Networks. In: Biological Networks, edited by François Képès, World Scientific: , pp. 1-.

R. Brette and 21 others (2007) Simulation of networks of spiking neurons: A review of tools and strategies. Journal of Computational Neuroscience 23: 349-398.

We review different aspects of the simulation of spiking neural networks. We start by reviewing the different types of simulation strategies and algorithms that are currently implemented. We next review the precision of those simulation strategies, in particular in cases where plasticity depends on the exact timing of the spikes. We overview different simulators and simulation environments presently available (restricted to those freely available, open source and documented). For each simulation tool, its advantages and pitfalls are reviewed, with an aim to allow the reader to identify which simulator is appropriate for a given task. Finally, we provide a series of benchmark simulations of different types of networks of spiking neurons, including Hodgkin-Huxley type, integrate-and-fire models, interacting with current-based or conductance-based synapses, using clock-driven or event-driven integration strategies. The same set of models are implemented on the different simulators, and the codes are made available. The ultimate goal of this review is to provide a resource to facilitate identifying the appropriate integration strategy and simulation tool to use for a given modeling problem related to spiking neural networks..
Supported by the European Community (FACETS project, IST 15879), NIH (NS11613), CNRS and HFSP.

[Preprint from arXiv] [get the models from ModelDB] [PubMed]
Davison A.P. and Frégnac Y. (2006) Learning crossmodal spatial transformations through spike-timing-dependent plasticity. The Journal of Neuroscience 26: 5604-5615.

A common problem in tasks involving the integration of spatial information from multiple senses, or in sensorimotor coordination, is that different modalities represent space in different frames of reference. Coordinate transformations between different reference frames are therefore required. One way to achieve this relies on the encoding of spatial information with population codes. The set of network responses to stimuli in different locations (tuning curves) constitutes a set of basis functions that can be combined linearly through weighted synaptic connections to approximate nonlinear transformations of the input variables. The question then arises: how is the appropriate synaptic connectivity obtained? Here we show that a network of spiking neurons can learn the coordinate transformation from one frame of reference to another, with connectivity that develops continuously in an unsupervised manner, based only on the correlations available in the environment and with a biologically realistic plasticity mechanism (spike timing-dependent plasticity).
Supported by EU project IST-2001-34712.

[get the model from ModelDB] [PubMed] [full text]
Badoual M., Zou Q., Davison A.P., Rudolph M., Bal T., Frégnac Y. and Destexhe A. (2006) Biophysical and phenomenological models of multiple spike interactions in spike-timing dependent plasticity. International Journal of Neural Systems 16: 79-97.

Spike-timing dependent plasticity (STDP) is a form of associative synaptic modification which depends on the respective timing of pre- and post-synaptic spikes. The biophysical mechanisms underlying this form of plasticity are currently not known. We present here a biophysical model which captures the characteristics of STDP, such as its frequency dependency, and the effects of spike pair or spike triplet interactions. We also make links with other well-known plasticity rules. A simplified phenomenological model is also derived, which should be useful for fast numerical simulation and analytical investigation of the impact of STDP at the network level.
Supported by EU project IST-2001-34712.

[PubMed] [full text]
Davison A.P. (2004) Biologically-detailed network modelling. In: Computational Neuroscience: A Comprehensive Approach, edited by J. Feng, Chapman and Hall/CRC Press: Boca Raton, pp. 287-304.

Davison A.P., Feng J. and Brown D. (2003) Dendrodendritic inhibition and simulated odor responses in a detailed olfactory bulb network model. Journal of Neurophysiology 90: 1921-1935.

In the olfactory bulb both the spatial distribution and the temporal structure of neuronal activity appear to be important for processing odor information, but it is currently impossible to measure both of these simultaneously with high resolution and in all layers of the bulb. We have developed a biologically-realistic model of the mammalian olfactory bulb, incorporating the mitral and granule cells and the dendrodendritic synapses between them, which allows us to observe the network behavior in detail. The cell models were based on previously published work. The attributes of the synapses were obtained from the literature. The pattern of synaptic connections was based on the limited experimental data in the literature on the statistics of connections between neurons in the bulb. The results of simulation experiments with electrical stimulation agree closely in most details with published experimental data. This gives confidence that the model is capturing features of network interactions in the real olfactory bulb. The model predicts that the time course of dendrodendritic inhibition is dependent on the network connectivity as well as on the intrinsic parameters of the synapses. In response to simulated odor stimulation, strongly activated mitral cells tend to suppress neighboring cells, the mitral cells readily synchronize their firing, and increasing the stimulus intensity increases the degree of synchronization. Preliminary experiments suggest that slow temporal changes in the degree of synchronization are more useful in distinguishing between very similar odorants than is the spatial distribution of mean firing.

[get the model from ModelDB] [PubMed] [full text]
Davison A.P. and Shepherd G.M. (2002) Olfactory Bulb. In: The Handbook of Brain Theory and Neural Networks, 2nd Edn, edited by M. Arbib, The MIT Press: Cambridge, MA.

Davison A.P., Morse T.M., Migliore M., Marenco L., Shepherd G.M. and Hines M.L. (2002) ModelDB: A Resource for Neuronal and Network Modeling. In: Neuroscience Databases: A Practical Guide, edited by R. Kötter, Kluwer Academic Publishers: Norwell, MA, pp. 99-109.

ModelDB is an online database (senselab.med.yale.edu/senselab/ModelDB) of published neuronal models, including models of ion channels, dendrites, axons, neurons, synapses and networks of neurons. Having ready access to the code for a model facilitates testing and verification of a model, re-use of model components to speed development of new models, and comparing a model to new experimental data. The database is useful for archiving models and for collaboration on modeling projects, and is a resource for teachers and students in both theoretical and experimental neuroscience. We describe here how to use the database: how to find specific models, how to obtain model code, how to run models, and how to contribute a model to the database.

Davison A.P. (2001) Mathematical modelling of information processing in the olfactory bulb. Ph.D. thesis, University of Cambridge.

The aim of this dissertation is to investigate the processing of sensory signals in the mammalian olfactory bulb, using analysis and computer simulation of mathematical models. A biologically-detailed mathematical model provides a framework which integrates the results of experiments at different levels of enquiry, and enables study of problems which cannot easily be addressed using only the methods of experimental neuroscience.
Specific biological and computational problems which are addressed include: the existence, origin and role of oscillations/synchronisation; how the properties of individual cells/synapses influence the network behaviour; the role of lateral inhibition; how the connectivity between cells influences network behaviour.
The dissertation has four main parts: (i) a review of the anatomy and physiology of vertebrate olfactory systems, and of previous modelling studies of the olfactory bulb; (ii) development of biophysical models of the principal neurone types of the olfactory bulb, based closely on experimental data, but simple enough to allow simulation of large networks; (iii) an examination of the fundamental interaction in the bulb -- that between two mitral cells -- using simulation of the biophysical cell models and analysis of the simpler integrate-and-fire neurone model; (iv) development of network models of the olfactory bulb incorporating the biophysical neurone models. These are tested using experimental data from the literature, and then the properties of the network are studied, leading to predictions which could be tested experimentally.

[PDF]
Davison A.P., Feng J. and Brown D. (2000) A reduced compartmental model of the mitral cell for use in network models of the olfactory bulb. Brain Research Bulletin 51: 393-399.

We have developed two-, three- and four-compartment models of a mammalian olfactory bulb mitral cell as a reduction of a complex 286-compartment model. A minimum of three compartments, representing soma, secondary (basal) dendrites and the glomerular tuft of the primary dendrite, is required to adequately reproduce the behaviour of the full model over a broad range of firing rates. Adding a fourth compartment to represent the shaft of the primary dendrite gives a substantial improvement. The reduced models exhibit behaviours in common with the full model which were not used in fitting the model parameters. The reduced models run 75 or more times faster than the full model, making their use in large, realistic network models of the olfactory bulb practical.

[get the model from ModelDB] [PubMed] [full text]

Conferences

Davison A.P., Yger P., Kremkow J., Perrinet L. and Muller E. (2007) PyNN: Towards a universal neural simulator API in Python.
CNS*2007, Toronto, July 2007.

Trends in programming language development and adoption point to Python as the high-level systems integration language of choice. Python leverages a vast developer-base external to the neuroscience community, and promises leaps in simulation complexity and maintainability to any neural simulator that adopts it. PyNN [http://neuralensemble.org/PyNN] strives to provide a uniform application programming interface (API) across neural simulators. Presently NEURON and NEST are supported, and support for other simulators and neuromorphic VLSI hardware is under development. With PyNN it is possible to write a simulation script once and run it without modification on any supported simulator. It is also possible to write a script that uses capabilities specific to a single simulator. While this sacrifices simulator-independence, is adds flexibility, and can be a useful step in porting models between simulators. The design goals of PyNN include allowing access to low-level details of a simulation where necessary, while providing the capability to model at a high level of abstraction, with concomitant gains in development speed and simulation maintainability. Another of our aims with PyNN is to increase the productivity of neuroscience modeling, by making it faster to develop models de novo, by promoting code sharing and reuse across simulator communities, and by making it much easier to debug, test and validate simulations by running them on more than one simulator. Modelers would then become free to devote more software development effort to innovation, building on the simulator core with new tools such as network topology databases, stimulus programming, analysis and visualization tools, and simulation accounting. The resulting, community-developed 'meta-simulator' system would then represent a powerful tool for overcoming the so-called complexity bottleneck that is presently a major roadblock for neural modeling..
Supported by the European Community (FACETS project, IST 15879).

Davison A.P. (2006) Simulator-independent network modelling with Python and XML.
Modeling the Brain's Labyrinth, Fodele, Crete, Greece, September 2006.

Reproducibility of experimental results is the foundation of science, and in this area computational neuroscience should have a huge advantage since in principle computational experiments can be reproduced exactly. In practice, reproduction is often difficult, in part because Methods sections of papers often give insufficient detail to recreate a model and the original code is not always made available, and in part because of the multiplicity of simulation environments that are available: conversion of a model from one simulation tool to another is rarely straightforward. In an attempt to address the latter problem I have developed an interface, PyNN (http://neuralensemble.org/PyNN), using the Python programming language, that allows a neuronal network model to be written once, then simulated on multiple simulators with no change in the model code. For simulators which already have a Python interface, PyNN controls the simulator directly. In other cases, PyNN writes native code which can then be run on the simulator in the usual way. Network connectivity and simulation control are written in Python, but for specification of individual neuron models PyNN takes advantage of the NeuroML standards. PyNN is certainly useful for developing new network models that are not dependent on a particular simulator, and so can be run in several different environments to reduce the occurrence of bugs and to find the most efficient simulator for a given model. It is also useful for converting a model from one simulator to another, since it is easier to convert a model from a specific simulation language to PyNN, which can use the same simulation engine, than directly to a different language/engine: the subtle differences between the two simulators that make conversion difficult have already been taken account of in PyNN and hence do not have to be considered in the conversion process. .
Supported by the European Community (FACETS project, IST 15879).

Davison A.P., Yger P., Chan J.S., Newell F.N. and Frégnac Y. (2006) A combined psychophysical-modelling study of the mechanisms of tactile picture perception.
CNS*2006, Edinburgh, U.K, July 2006.

The involvement of visual cortex in haptic spatial perception, including haptic object recognition, is well established, but little is known about the precise mechanisms involved. We have pursued a joint experimental-modelling approach to explore possible mechanisms. Human exploratory paths in a task involving tactile picture perception are used to drive a system-level model of haptic processing, based on the idea that humans continuously compare tactile and proprioceptive sensory input to multiple hypotheses about the contents of the picture. Each hypothesis is represented as a mental image, expressed as a pattern of activation in visual cortex representing the expectation of finding particular tactile features at particular points in space. The simulated responses of the model show good agreement with human performance when averaged over subjects, but fit less well on a subject-by-subject basis.
Supported by EU project IST-2001-34712.

Davison A.P. (2006) NModLib: a database of components for neuroscience models.
World Association of Modeling - Biologically Accurate Modeling Meeting, San Antonio, Texas, March 2006.

NModLib is a database containing NMODL mechanisms, useful hoc functions and templates and other add-on components for the NEURON simulator. It is complementary to the ModelDB database, but where the focus of ModelDB is on complete models, the focus of NModLib is on model components, i.e. a place to find mechanisms for incorporation into your own model and avoid reinventing the wheel. In addition, ModelDB is limited to models that have been described in a peer-reviewed publication. This restriction does not apply to NModLib. Most of the NMODL files in ModelDB are also available in NModLib, with links back to the full models of which they are part.

Davison A.P., Belatreche A., Fieres J., Chan J.S., Newell F.N. and Frégnac Y. (2005) The Two-Ring paradigm: a computational approach to multisensory code binding.
International SenseMaker Workshop on Life-Like Perception Systems, Derry, U.K, April 2005.

The Two-Ring Problem is a framework for investigating the integration of multisensory information in humans and in a machine (the SenseMaker System), specifically visual and haptic information. The tasks to be attempted in the Two-Ring Problem involve the perception of static, two-dimensional scenes: the machine or observer has to segment a scene into its component objects, perceptually identify those objects and infer their relative spatial relationships. The tasks may be attempted using vision (an image is presented on a screen), the haptic sense (a finger is scanned over a textured surface), or both senses together. Comparing human and machine performance under different constraints will allow us to explore, using computational methods, neural-based hypotheses for the perceptual mechanisms underlying this performance. In this paper we outline the biology-inspired architecture and implementation of the neuronal-network model that will run on the SenseMaker system hardware, in order to solve multi-sensory perception tasks in the Two-Ring framework.
Supported by EU project IST-2001-34712.

Davison A.P. and Frégnac Y. (2004) Training a network of spiking neurons to perform coordinate transformations. Society for Neuroscience Abstracts 30: 177.4.
Society for Neuroscience Annual Meeting, San Diego, California, November 2004.

Davison A.P. (2004) Learning simple computations in spiking basis function networks using spike time-dependent plasticity. FENS Forum Abstracts 2: A224.4.
FENS, Lisbon, Portugal, June 2004.

Spatial variables, such as the location of a stimulus in space, are commonly represented in the brain by population codes. Denève, Latham and Pouget (2001; Nature Neurosci 4:826-831) have shown that basis function networks can perform a variety of computations on population-encoded variables. This approach has proven to be very successful and could, for example, be used to perform coordinate transformations. However, the model was based on firing-rate representations of neuronal activity, and the synaptic weights were either calculated or learned using classical neural network learning algorithms such as the delta rule. In order to make the approach more biologically realistic, we implemented models of basis function networks with spiking neurons and with spike timing-dependent synaptic plasticity. These models are able to learn the connection weights required to compute coordinate transformations using population codes, in a manner equivalent to rate-based models but with a greater degree of biological realism. These results support the hypothesis that the brain may indeed use such principles for performing computations.
Supported by EU project IST-2001-34712.

Davison A.P., Hines M.L. and Shepherd G.M. (2002) Membrane bistability and sub-threshold oscillations in an olfactory bulb mitral cell model.
AChemS XXIV, Sarasota, Florida, April 2002.

Davison A.P., Zhou Z., Hines M.L. and Shepherd G.M. (2001) Simulating sodium and potassium currents in an olfactory mitral cell model.
Society for Neuroscience Annual Meeting, San Diego, California, November 2001.

We have previously developed a tightly constrained model of the olfactory mitral cell based on dual patch recordings in rat olfactory bulb slices (Shen et al, J. Neurophysiol. 82: 3006-3020, 1999). The model closely simulates action potential generation at somatic and dendritic sites based on a transient Na and delayed rectifier K current. In order to extend this model for other membrane currents, we have carried out a combined experimental (patch-clamp) and modeling study. The response to long-duration, perithreshold current injection from a hyperpolarized resting potential has three main characteristics: (i) a delay of up to several hundred milliseconds before action potential firing; (ii) an inflection or overshoot in the membrane potential rise phase; (iii) subthreshold membrane potential oscillations. With suprathreshold stimulation the firing frequency increases with time, i.e. negative adaptation. These characteristics suggest the presence of a slowly-inactivating K current and a persistent Na current. We are extending the previous model in order to test these hypotheses by characterizing the membrane currents and analysing the contribution each makes to the observed behaviour. Channel kinetics and channel density distribution are obtained by fitting the model to the experimental recordings. Channel kinetics are also obtained by voltage clamp recordings and there is good agreement between the parameters obtained by the two methods. The results will lead to simulations of the firing patterns of the mitral cell over time, so that models of synaptic activation and mitral-granule cell networks can be constructed.
Supported by NIDCD and NIMH (Human Brain Project).

Davison A.P. and Feng J. (2001) A model of network interactions in the olfactory bulb.
AChemS XXIII, Sarasota, Florida, April 2001.

We have developed a detailed, biologically-realistic model of the mammalian olfactory bulb, incorporating the mitral and granule cells and the dendrodendritic synapses between them. The individual cell models were simplified from detailed compartmental models which had been fitted to experimental data. The amplitudes, time courses and transmission delays of the synapses were obtained from the literature. A simple method for specifying the synaptic connections was adopted, based on the limited experimental data in the literature on the statistics of connections between neurons in the bulb. Both electrical and odor stimulation were modeled. A simple model of olfactory inputs was used which captures some qualitative aspects of odor inputs but which is not necessarily quantitatively accurate. As a test of the model, a series of simulation experiments with electrical stimulation were performed and the results agreed quite closely with published experimental data which were not used in developing the model. This gives confidence that the model is capturing some features of network interactions in the real olfactory bulb. Simulation experiments with `odor' stimulation were then performed to investigate: (i) how the model response (in terms of synchronization and the spatial distribution of activity) is affected by stimulus intensity; (ii) how the response depends on connectivity parameters; and (iii) whether the network makes it easier to discriminate between similar odor inputs.

Davison A.P., Feng J. and Brown D. (2001) Spike synchronization in a biophysically-detailed model of the olfactory bulb. Neurocomputing 38-40: 515-521.
CNS*00, Brugge, Belgium, July 2000.

Stimulus-evoked synchronization of action potentials has been demonstrated in mammalian olfactory bulb and in insect antennal lobes. Abolition of synchronization has been shown to impair the ability of honeybees to perform fine olfactory discrimination. We present a biophysically-detailed computer model of the olfactory bulb which qualitatively reproduces many features seen in experimental recordings. The mitral cells of the model synchronize readily without common input due to lateral interactions with inhibitory granule cells. Weakly activated mitral cells fire more slowly than, but always synchronously with, strongly activated cells. Nearby cells synchronize more readily than widely-separated ones.

[full text]
Davison A.P., Feng J. and Brown D. (1999) Structure of lateral inhibition in an olfactory bulb model. Lecture Notes in Computer Science 1606: 189-196.
IWANN'99, Alicante, Spain, June 1999.

It has been shown that mutual lateral inhibition of the projection neurones in the olfactory bulb, mediated by interneurones, serves to tune the representation of odours in the bulb and reduce the overlap between similar odorants. In this paper we demonstrate that the parameters of the lateral interaction, specifically the relation of synaptic strength to cell separation and the effective overall gain of the network, have a significant effect on the strength and range of lateral inhibition in a simple model of the olfactory bulb.

Davison A.P., Miller T., Belton I., Bolton S. and Bonnett D.E. (1997) An assessment of image registration in the treatment planning of tumours of the brain.
Radiology 1997 - Imaging, Science and Oncology, Birmingham, U.K, May 1997.