This action funds an NSF Postdoctoral Research Fellowship for FY 2010. The fellowship supports a research and training plan entitled "Understanding massively parallel neural computations: next-generation analysis of simultaneous recordings from thousands of neurons" for Peter Li. The host institution for this research is Salk Institute for Biological Studies, and the sponsoring scientist is E.J. Chichilnisky.
Understanding the massively parallel computations of the nervous system requires high-resolution techniques for monitoring the activity of many neurons in a circuit simultaneously, combined with mathematical approaches to analyze and interpret the results. Recent advances in technologies allow simultaneous recording of thousands of retinal ganglion cells (neurons in the eye that process information about the visual world and transmit that information to the brain). This project implements next-generation computational analyses for massively parallel neural recordings. For example: how do the many known subtypes of retinal ganglion cells work in parallel? Technical challenges include large volumes of data and high computing demands. Thus, one component of the project is leveraging high-performance supercomputing clusters, both to allow new analyses that would be impractical on standard computers, and to allow efficient deployment of analyses over this large (~100 TB) existing data archive.
Training goals for this project include building mathematical knowledge in systems analysis, functional analysis, and information theory, which are valuable skills in today's biological research fields. Understanding how vision works would be a singular triumph as well as a major stepping stone towards the greater goal of unlocking the mysteries of the brain. Valuable knowledge is being gained by comparing human vision and cognition to the abilities and specializations of other species.
The retina is a part of the nervous system that sits at the back of the eye and catches light, converting optical signals into electrical signals that are sent to the brain. But the retina is not simply like the pixels in a digital camera; it processes the visual information in fundamental ways before passing it on to the brain, performing behaviorally relevant functions such as edge detection and contrast enhancement. In this project, we developed new techniques for stimulating the retina with light at a very fine spatial scale and recording its responses, allowing us to study visual processing at the unprecedented resolution of elementary inputs and outputs. Thus we were able to see what information gets passed to the brain when we stimulated a single retinal input cell ( AKA cone photoreceptor), or when we stimulated multiple retinal inputs in specific combinations. These experiments revealed fundamental things about retinal processing that are also relevant for the understanding of general neural processing. For example, how do the signals from multiple cone inputs combine before being sent to the brain? Is the response of the retina to stimulation of two cone photoreceptors in combination simply equal to the sum of the responses to each cone stimulated alone? Using our new high resolution approach, we could test exactly this type of question, and we found that under some specific conditions the simple sum applies, but in many cases the retina does something more complicated. We then used our fine control of retinal stimulation to help deduce what kinds of reitnal circuits might be able to produce these different behaviors. To further our understanding of how the structure of the retinal circuitry determines its functional behavior, this project also funded a new approach to combining electrical recording of the retina with anatomical staining and genetic testing at a cellular level. By combining our recordings of thousands of retinal neurons with cellular stains specific for certain gene products (i.e. proteins) we were able to match cells with known electrical behavior with a particular anatomical and genetic profile. So, for example, we were able to draw conclusions such as "retinal cells that respond to brightness have this gene and this shape, while retinal cells that respond to darkness have this other gene and this other shape." Although the funding period has ended, both these projects are continuing on. We have produced several conference presentations based on this work and we are producing several manuscripts for submission to peer-reviewed journals. Making these discoveries required considerable computational infrastructure. This project funded many improvements to stimulation and analysis software, for example using multiple computers connected in a cluster to speed analysis and allow us to react more quickly if the retina responded to our experimental stimulation in unexpected, important, and exciting ways. Furthermore, some of the software improvements and techniques developed as part of this project, have been released to the public as open source and online articles so that other researchers can benefit. Being funded by this project gave me the opportunity to learn a great deal about high performance software development, statistical analysis of large, high-dimensional datasets, anatomical staining of fixed tissue, and mammalian physiology. In the course of working on this project, I also had the opportunity to train several colleagues in these methods, especially drawing on my experience in software development and anatomy, and I gained valuable experience in managing small groups to achieve a common goal.
