Computational neuroscience and the NeuroHub project
Figure 1 (Click image to enlarge) The locust hind leg control system and the sensory neurones, interneurones and motor neurones and muscle.
Computational neuroscientists study nervous system function in terms of the information processing properties of nerve cells, or neurones, that form neural networks. We aim to develop models to integrate large and complex experimental datasets to understand brain function. By its very nature computational neuroscience is an area of research that can be truly interdisciplinary, bringing together the fields of neuroscience with engineering, computer science, mathematics and physics, in addition to many other biological disciplines. Rarely is all the expertise required to understand computational problems in the same country let alone the same department or University, so there is a growing need to share data, techniques and expertise at a global level (Teeters et al., 2008) to ultimately reduce experiments on animals and to improve understanding of brain function. Large scale international collaborations require new approaches to research, mechanisms of coordination and funding and the appropriate infrastructure to achieve those goals.
Partly driven by research council data sharing policies, for example the UK Medical Research Council (2006), research journal requirements for data accessibility, and partly by a deeper understanding of the benefits of data sharing learnt form other research fields (Insel et al., 2003), there is a significant shift amongst the neuroscience community to embrace an exchange of neuroscience data and tools. Molecular biology, proteomics and genomics lead the way in the sharing of primary data and infrastructure, resources and policies that promote the sharing of data (Kay et al., 2009), with the key aim of making data available in the public domain to encourage research and development (The Bermuda Principles, 1996). As a result significant strides have been made in the use of such resources, ultimately maximising its benefits to society and enabling complex genome-wide associations and gene–environment interaction studies to be performed. Clearly there are many challenges with such an approach in neuroscience, however many of the ground rules have been established, good and bad practices developed, the benefits and pitfalls defined (Foster and Sharp, 2007) and all from which the neuroscience community can learn.
Not withstanding the many issues surrounding the collection of neurobiological data such as the need for standardising data formats, using standardised inputs, common models (Nordlie et al., 2009), and image formats, to mention only a few, there has been the key issue surrounding the best infrastructure to support the community. As neuroscientists do we simply require a repository for data that can be accessed via the internet, and that may satisfy a small cohort of scientists? If so then there are many solutions available, but if we want to go further to establish a virtual research environment that can be used globally, to allow us to work within an accessible research environment that integrates across the many fields within the discipline (molecular, genetic, physiological, anatomical, behavioural) then we need to look for something new.
The NeuroHub project funded by Joint Information Systems Committee (JISC A2 strand developing e-infrastructure to support research disciplines) aims to address this gap by developing a set of sustainable tools and a framework that allows the neuroscience community to efficiently and effectively use existing e-Infrastructure. The project aspires to change the way in which we, neuroscientists, carry out our research by identifying the key challenges and providing the software infrastructure that provides efficient and effective use of key e-Infrastructure to manipulate, analyse and share data. By doing so the project enables a more productive research cycle, from conception of experiment to publication of the research results. Different neuroscientists require different tools and hence the framework and tools of NeuroHub are the product of in-depth user requirements discussions with neuroscientists and computer scientists working closely together, adaptation of existing software, development of key or missing components and a tight collaboration between a cross-domain team of neuroscientists, computer scientists, technologists and resource providers in Southampton, Oxford and Reading. The project supports the entire research lifecycle from investigation of prior knowledge, through experimentation, computation, analysis and dissemination of results.
In Southampton Phil Newland’s research group (funded by the BBSRC and EPSRC) studies how the nervous system produces limb movements. Insects produce very precise movements of their limbs even though the numbers of neurones within the nervous system are many orders of magnitude less than in the human brain. We ask how this remarkable control is achieved and how the relatively small numbers of neurones in the insect nervous system achieves the high degree of versatility of movement and similar precision of higher animals. We know little of how the nervous system performs the complex integrative task of processing different sensory parameters; are there specialist neurones or populations of neurones that process different signals (high frequency versus low frequency inputs, or velocity versus position or acceleration), what are the benefits of digital (spike-train) signalling compared to analogue (continuously varying cell potentials) signalling, and what do the different branches of a neurone contribute to integration compared to the whole cell? Understanding how the insect nervous system solves the problem of controlling precise movements will ultimately have many benefits, including a better understanding of how the nervous systems performs complex control tasks, generating new hypotheses on how the brains of more complex vertebrates solve similar tasks, and will provide the basis for biologically inspired engineering solutions to movement control. The work is being carried out in collaboration between Phil Newland in the Centre for Biological Sciences and David Simpson, Robert Allen, Brian Mace and Emiliano Rustighi from the Institute for Sound and Vibration Research at the University of Southampton. The need for global access is highlighted in our research group through international collaboration with Dr Xingjian Jing in Hong Kong and Prof Carlos Maciel in Brazil, whose expertise in mathematics and Information Theory, respectively, provide additional approaches to understanding network function.
How is NeuroHub likely to help us in our a research? Our research focuses on the control of movements about the ‘knee’ joint of the locust. We stimulate a key sensor in the limb, the femoral chordotonal organ (FeCO), with band-limited Gaussian White Noise (the input) and record the activity of individual neurones in the control network; including sensory neurones, interneurones that communicate digitally with nerve impulses or via analogue signalling, motor neurones and muscle activity and force output (the output) (see Figure1). In this way we build up a complete picture of the input-output properties of all of the different components in the neural networks and the muscle outputs.
Having collected enormous datasets there is a need for archiving and curation, searching archives for neurones of a particular type for analysis, and then performing the analysis itself. Archiving and searches of large datasets can be best achieved efficiently and quickly through a digital environment such as NeuroHub, while sharing of analytical tools and results is where NeuroHub will be truly invaluable. To quantify the responses of the neurones in different layers of a network we use a variety of signal processing techniques including, for example, the Wiener kernel method to establish nonlinear models to characterise the input sensitivity of specific populations of interneurones ( Vidal-Gadea et al., 2009). Different algorithms can be used to define the models and each can be tested, models and results posted on NeuroHub, and online discussion of results carried out across the research group. In this way we often perform multiple analyses on the same data sets, and as new tools or algorithms become available in the signal processing field they can also be tested on the same datasets. With an infrastructure that allows continuous uploading of linked files NeuroHub provides a complete time-stamped history of the data and all analysis, finally culminating in the publication itself.
Importantly, the NeuroHub infrastructure allows the integration of the many different fields in neuroscience. For example while our computational analysis reveals the response patterns of individual groups or populations of neurones, our experiments demonstrate the effect a particular neurone has on behaviour. Thus behaviour is recorded and the data linked to a particular physiological experiment in NeuroHub. Finally we need to identify individual neurones and we do that by injecting them with a fluorescent dye that is imaged using high resolution confocal or wide field microscopy. This information is also linked within NeuroHub so that a complete record of a neurone’s morphology, physiology and integrative properties are linked to computational models and resultant behaviour that can be generated. The data can be made public thus maximising its use in developing computational tools or generating a greater understanding of network function that will ultimately inform at different levels, from biology to engineering.
Links and References
Foster MW, Sharp RR (2007). Share and share alike: deciding how to distribute the scientific and social benefits of genomic data. Nature Rev Genetics. 8: 633-639.
2003) Neuroscience Networks: Data-sharing in an information age. PloS Biol 1(1): e17.
Kaye J, Heeney C, Hawkins N, de Vries J, Boddington B. (2009). Data sharing in genomics - re-shaping scientific practice. Nature Rev Genetics. 10: 331-335.
Nordlie E, Gewaltig M-O, Plesser HE (2009). Towards reproducible descriptions of neuronal network models. PLoS Comput Biol. 5(8): e1000456.
Teeters JL, Harris KD, Millman KJ, Olshausen BA, Sommer FT (2008). Data sharing for computational neuroscience. Neuroinform 6(1): 47-55.
The Bermuda Principles (1996): http://www.ornl.gov/sci/techresources/Human_Genome/research/bermuda.shtml#1
UK Medical Research Council (2006): http://www.mrc.ac.uk/Ourresearch/Ethicsresearchguidance/Datasharinginitiative/Policy/index.htm
Vidal-Gadea AG, Jing XJ, Simpson D, Dewhirst OP, Kondoh Y, Allen R, Newland PL (2010). Coding characteristics of spiking local interneurons during imposed limb movements in the locust. J. Neurophysiol. 103: 603-615.
The NeuroHub Team: Anne Trefethen, David Wallom, Nigel Emptage, Neil Caithness, Steven Young, John Pybus, Kang Tang, David Spence (University of Oxford); David De Roure, Jeremy Frey, Vincent O'Connor, Philip Newland, David Newman (University of Southampton); Mark Baker, Douglas Saddy, Garry Smith (University of Reading).