Neuroinformatics

Phi Kappa Phi Forum, Winter 2005 by Arbib, Michael A

INTRODUCTION

The key level of analysis of the brain is the neuron - a cell specialized to respond to complex patterns of activity from receptors (the body's sensors) or other neurons both to change its state (this provides the key to memory) and also to send a timed pattern of signals to other neurons and then through the complexity of networks of neurons to the glands and the muscles (the body's effectors). The point at which one neuron connects to another is called a synapse. The synapse is a complex chemical machine, but so too is the cell itself; understanding gene expression is a crucial part of understanding this machine. The human brain has hundreds of anatomically and functionally distinct regions. Among them, these regions have perhaps as many as one-hundred-billion neurons, and some neurons have tens of thousands of synapses.

Just to add one more daunting number, the October 2004 meeting of the Society for Neuroscience (SfN) drew more than thirty-thousand people to the San Diego Convention Center, and almost all of them are active researchers, adding several papers a year to the literature on the brain. In short, the brain is an incredibly complex system, and responses to the challenge of understanding its functions and mechanisms produce a vast torrent of both new empirical data and publications presenting that data.

FROM GENES TO BRAINS

Neuroinformatics, then, might be defined simply as the use of computers (information technology) to assist neuroscientists in managing that torrent. However, the use of computers has become so ubiquitous in the developed world that we need a more restrictive focus. One key element that must be included is the extensive use of databases, as was set forth in the deliberations of the Committee on a National Neural Circuitry Database of the Institute of Medicine (Mapping the Brain and Its Functions: Integrating Enabling Technologies into Neuroscience Research, Constance Pechura and Joseph Martin, Editors, National Academy Press, Washington, D.C., 1991). Although that report used the word "database" hundreds of times and mentioned the concept of simulation only in passing, it is now widely accepted that neuroinformatics goes well beyond the use of databases, the World Wide Web, and visualization in the storage and analysis of neuroscience data. The structuring of masses of data by a variety of computational models is essential to the future of neuroscience. From this perspective, neuroinformatics includes computational neuroscience, the use of computational techniques and metaphors to investigate relations between neural structure and function.

In recent years, the Human Genome Project has become widely known for its success in sequencing the complete human genome and providing the results in comprehensive databases such as GenBank, the National Institutes of Health (NIH) genetic sequence database, which is an annotated collection of all publicly available DNA sequences. It contained approximately thirty-eight billion bases in thirty-three million sequence records as of February 2004. However, the key data are rather simple, taking the form of annotated base-pair sequences of DNA.

In the hope of doing for neuroscience what the Human Genome Project has done for human genetics and inspired by the report Mapping the Brain and Its Functions described earlier, NIH announced the Human Brain Project (HBP) on April 2, 1993, with the goal of providing a World Wide Web-based set of neuroscience databases interoperable with each other as well as with genomic and other databases. However, by contrast with the single-data format that allows GenBank to rapidly absorb millions of new entries every year, neuroinformatics databases must be built for immensely heterogeneous sets of data. Not only must one contend with the many orders of magnitude linking the finest details of neurochemistry to the overall behavior of the organism, but also one must integrate data gathered by many different specialists. Neuroanatomists characterize the brain's connectivity patterns; neurophysiologists characterize neural activity and the "learning rules" that summarize the conditions for, and dynamics of, change; neurochemists seek the molecular mechanisms that yield these "rules"; computational neuroscientists seek to place all these within a systems perspective.

To give some sense of the scope of neuroinformatics, I will briefly survey the contents of a book I coedited that describes the first five years of work of the University of California Brain Project (USCBP) on neuroinformatics (Arbib, M.A., and Grethe, J., Eds., Computing the Brain: A Guide to Neuroinformatics, San Diego: Academic Press, 2001) and then give some sense of the databases and resources made available via the Neuroscience Database Gateway of the Society for Neuroscience.

Figure 1 suggests the diversity of databases that need to be interconnected to form a federation which provides access to many forms of data, with illustrative material from work at the University of Southern California. The time-series data shown here are behavioral records (top traces), firing patterns of an individual neuron (middle traces), and a histogram (bottom) from studies of classical conditioning in the laboratory of Richard Thompson. The picture at top, from the laboratory of Michel Baudry, shows staining of a slice of brain for substances related to memory mechanisms; we call it atlas-based because interpretation requires registration of the slice against a standard atlas of the rat brain. At right, we are reminded of the wealth of information (such as published articles) distributed across the World Wide Web, and the challenges of "mining" and managing this overwhelming body of data. Finally, at bottom, the schematic of a model by Michael Crowley and me of the basal ganglia (disorders of which include Parkinson's disease and Huntington's disease) indicates the importance of theories and models needed to structure the manifold data of neuroscience. The key point, of course, is that data from very different sources will need to be brought together, thus challenging neuroinformaticians to do far more to make their resources interoperable.