Model sharing in computational neuroscience
Why model sharing is beneficial
Hallmarks of the scientific method are reproducibility and testability. It is crucial that experiments which test hypotheses can be performed (especially in other laboratories than they came from) and can be extended to test their ranges of validity. Computational neuroscience models (CNS models) are implicitly a collection of hypotheses of the function of (usually a subset of) the nervous system across spatial scales ranging from molecules to animal behavior. Most publications of CNS models describe how a computer program which faithfully imbued the collection of hypotheses behaves. The complexity of computational neuroscience models make it difficult to reproduce a model de novo from the publication that introduces it. Missing information and typographical errors are among the most common reasons why an investigator may fail to successfully reproduce a model.
The software implementation of a model can also become the building block(s) for modified versions of that model by other researchers, such as simplifying it for use in a larger spatial scale, including the effect of a variable (e.g. temperature on ion channel kinetics) that was absent in the original model, or extending the range of applicability of the model.
The media which stores computer code also obeys the laws of entropy; hard drives crash and old computers are recycled. Thus it is an important step to get the computer code into a database or onto some other maintained system to provide long-term preservation.
An analogy exists between the sharing of sequence information in sequence databases in traditional bioinformatics (the field comprised of sequence oriented databases, tools to search for sequences in one or more databases, compare with statistical software, and model the molecules related to the sequences) and the sharing of models in the CNS subset of neuroinformatics (the field comprised of storing, searching, comparing, and modeling diverse neuroscience data). Both types of sharing are essential to support future research in their respective fields.
Who is doing it now
Efforts to build databases to share CNS models have been underway since circa 1996 (Peterson et al. 1996, Mirsky et al. 1998, Usui 2003, Migliore et al. 2003, Sivakumaran et al. 2003, Hines et al. 2004, Le Novere et al. 2006) and also an alternative approach for sharing models in a napster-like method has appeared (Cannon et al. 2002).
Sites that support the sharing of CNS models:
- ModelDB: http://senselab.med.yale.edu/modeldb
- Visiome: http://platform.visiome.neuroinf.jp/
- CellML http://www.cellml.org/
- Cellular Open Resource (COR) http://cor.physiol.ox.ac.uk/
- BioModels Database: http://www.ebi.ac.uk/biomodels/
- Systems Biology Markup Language (SBML) http://sbml.org/
- Database of Quantitative Cellular Signaling (DOQCS): http://doqcs.ncbs.res.in
ModelDB and Visiome started out with particular domains of interest, olfaction and vision, respectively, and have subsequently expanded to broader categories of neuroscience and biological mathematical models. CellML has always been intended as a generic way to represent mathematical models, although, as the name implies, it was initially developed by groups with an interest in representing biological models, and its use in practice has been for biological mathematical models. The COR converts CellML models into executable models on the windows platform. BioModels contains diverse models: models of cell cycle, bacterial growth, plant photosynthesis, basic metabolism etc. The neuroscience models are a minority of their entries. The BioModels database contains models stored as SBML files. SBML is currently the most widely supported model definition standard, with more than 100 compatible applications listed on the SBML home page. The DOQCS is focused on the signaling pathways of the glutamate synapse.
Current and future directions
It is expected that the above resources will continue to grow and have increased functionality and interoperability with journals and the National Library of Medicine (PubMed). One important area that is the focus of much current work is the construction of "simulator independent" descriptions of models. The objective is to separate the model from a particular software implementation which then allows modelers to use models from other simulators within their preferred simulator. This work is based in XML and can be further examined at the following:
- Systems Biology Markup Language (SBML) http://sbml.org/index.psp
- CellML http://cellml.org/
- XML for computational neuroscience (NeuroML) http://www.neuroml.org/
- XML for Neuronal Morphology Data: http://www.morphml.org/ and http://www.morphml.org:8080/NeuroMLValidator/
- NeuroML Interfaces for GENESIS 3: http://www.genesis-sim.org/GENESIS/G3
- ChannelDB: http://www.genesis-sim.org/hbp/channeldb/ChannelDB.html
SBML and CellML (see their descriptions above) have many citations (36 hits for sbml and 12 hits for CellML in pubmed on April 3rd, 2007, see references for links) and a conversion program between them. Collaborations between the SBML and CellML groups have created prototype standards for sharing computational models (see Box 3 in Le Novere et al. 2005). NeuroML and it's morphology specific subset MorphML are tackling the complexity of the distribution of channels and receptors within neuronal morphology respectively (see Crook et al. 2007 and below links in references). The NeuroML validator allows automatic conversion of NeuroML files to NEURON and GENESIS formats. [neuroConstruct], a graphical application for the creation, visualization and analysis of network models, has just appeared (Gleeson et al. 2007). neuroConstruct's models are based on the NeuroML standards. [GENESIS] and [NEURON] developers have plans to develop and adopt XML descriptions of channel, cell, and network models. ChannelDB is a prototype of a Java-based NeuroML to GENESIS translator and database interface for ionic conductance models. More recent developments in this representation for conductance models have been carried out as ChannelML (a subset of NeuroML), and within the Neuroconstruct project.
See references for papers on XML related to CNS models.
Another approach to interoperability simply requires that the model be able to communicate with other models during the simulation (run-time interoperability, see Cannon et al. 2007 for a review of interoperability of neuroscience models, and Diesmann and Gewaltig 2002 for a run-time interoperability example).
The author would like to thank (but not hold responsible for errors) the kind help of Dave Beeman, Hugo Cornelis, Andrew Miller, Tommy Yu, and Nicolas Le Novere.
- Cannon RC, Gewaltig MO, Gleeson P, Bhalla US, Hines ML, Howell FH, Muller E, Stiles JR, Wils S, De Shutter E (2007) Interoperability of neuroscience modeling software: current status and future directions. Neuroinformatics 5(2):127-38
- Cannon RC, Howell FW, Goddard NH, De Schutter E.(2002) Non-curated distributed databases for experimental data and models in neuroscience. Network. 13(3):415-28. Review.
- Crook S, Gleeson P, Howell F, Svitak J, and Silver RA (2007) MorphML: Level 1 of the NeuroML standards for neuronal morphology data and model specification. Neuroinformatics 5(2):96-104
- Diesmann M, Gewaltig MO (2002) NEST: An Environment for Neural Systems Simulations in Forschung und wissenschaftliches Rechnen, GWDG-Bericht, pp 43-70. Plesser T, Macho V (eds), Göttingen (D):Ges. für Wissenschaftliche Datenverarbeitung.
- Gleeson P, Steuber V, Silver RA (2007) neuroConstruct: A tool for modeling networks of neurons in 3D space. Neuron 54:219-235
- Hines ML, Morse T, Migliore M, Carnevale NT, Shepherd GM (2004) ModelDB: A Database to Support Computational Neuroscience. J Comput Neurosci. 17(1):7-11.
- Le Novere N, Finney A, Hucka M, Bhalla US, Campagne F, Collado-Vides J, Crampin EJ, Halstead M, Klipp E, Mendes P, Nielsen P, Sauro H, Shapiro B, Snoep JL, Spence HD, Wanner BL (2005) Minimum information requested in the annotation of biochemical models (MIRIAM). Nat Biotechnol. 23(12):1509-15.
- Le Novere N, Bornstein B, Broicher A, Courtot M, Donizelli M, Dharuri H, Li L, Sauro H, Schilstra M, Shapiro B, Snoep JL, Hucka M (2006) BioModels Database: a free, centralized database of curated, published, quantitative kinetic models of biochemical and cellular systems. Nucleic Acids Res. 34(Database issue):D689-91.
- Migliore M, Morse TM, Davison AP, Marenco L, Shepherd GM, Hines ML (2003) ModelDB: making models publicly accessible to support computational neuroscience. Neuroinformatics 1(1):135-9.
- Mirsky JS, Nadkarni PM, Healy MD, Miller PL, Shepherd GM (1998) Database tools for integrating and searching membrane property data correlated with neuronal morphology. J Neurosci Methods 82(1):105-21.
- Peterson BE, Healy MD, Nadkarni PM, Miller PL, Shepherd GM (1996) ModelDB: an environment for running and storing computational models and their results applied to neuroscience. J Am Med Inform Assoc. 3(6):389-98.
- Sivakumaran S, Hariharaputran S, Mishra J, Bhalla US (2003) The Database of Quantitative Cellular Signaling: management and analysis of chemical kinetic models of signaling networks. Bioinformatics. 19(3):408-15.
Related to XML and model sharing
- SBML in pubmed http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?CMD=Display&DB=pubmed&term=sbml
- CellML in pubmed http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?CMD=Display&DB=pubmed&term=cellml
- NeuroML [publications]
- Tomasz Downarowicz (2007) Entropy. Scholarpedia, 2(11):3901.
- James M. Bower and David Beeman (2007) GENESIS. Scholarpedia, 2(3):1383.
- Marc-Oliver Gewaltig and Markus Diesmann (2007) NEST (NEural Simulation Tool). Scholarpedia, 2(4):1430.
- Ted Carnevale (2007) Neuron simulation environment. Scholarpedia, 2(6):1378.