Nerve cells in the brain extend long projections that twist and turn to form an intricate mesh. The points of contact between nerve cells, called synapses, are where the cells relay chemical signals to communicate with one another. Obtaining a map of the network of connections in the brain, or connectome, will be important to understanding complex brain functions, such as learning and memory, and even how certain diseases arise.
But mapping these billions of connections is a daunting task. Viren Jain is developing tools that will enable computers to do the process much more quickly than a purely human effort ever could.
One approach to mapping connectomes involves two steps: acquiring high-resolution image data and then analyzing the images. Scientists can obtain such image data by taking extremely thin slices of brain tissue and then zooming in about one hundred thousand times with an electron microscope (EM) to take images of the nerve cells. Vast amounts of image data are then analyzed to trace the path of every nerve cell and its projections and pinpoint all the connections they make, thereby generating a map of the neuronal connectivity in that brain tissue.
Using this approach, about 30 years ago, scientists mapped the connectome of the roundworm Caenorhabditis elegans, an organism that contains 300 neurons and 7,000 connections. With C. elegans scientists analyzed the EM images by hand—the project took 15 years to complete. But "to scale that approach to larger organisms, with hundreds of thousands to billions of neurons would be impossible unless computers do most of the work," says Jain. "We need to use computers to replace or at least facilitate what people can do."
The concept of instructing computers to perform useful tasks for people has fascinated Jain ever since he can remember. This interest led him to studying computer science as an undergraduate at the University of Pennsylvania. Then, in his last year of studies, his interest was piqued by neuroscience. "At that time, systems neuroscientists were starting to take measurements that described the functional properties of entire neural circuits and were able to relate those measurements to the behavior of the organism," he recalls. "I thought these advances were really interesting, so I decided to try to merge my interest in computation with neuroscience."
Jain asked his undergraduate research advisor at Penn, Lawrence Saul (who is now at the University of California, San Diego), whether he knew someone who was combining the field of machine intelligence with that of biological intelligence. Saul suggested the lab of Sebastian Seung at the Massachusetts Institute of Technology (MIT) in Boston. After joining Seung's lab, Jain spent a year learning various aspects of computational neuroscience. Seung then decided to switch the focus of the lab to solving the problem of automating image analysis for studying connectivity in neural circuits. Although at first skeptical of this ambitious project due to the difficulty of the problem, Jain says "eventually I became seduced by how fundamental a piece of the puzzle a connectome might be. So I worked on that problem for the next five years."
Jain and his colleagues at MIT developed an approach that enables computers to trace nerve cells from one EM image to the next with much higher accuracy than previous methods had been able to do.
After completing his Ph.D., Jain decided to join the Janelia Farm Research Campus to continue this line of work. "Janelia was in some sense the obvious choice. There is a clear investment in trying to understand neural circuits at multiple levels, and there are several labs developing new types of microscopy and analysis methods for doing so," says Jain. "Moreover, for someone who is doing research that could be looked at as slightly more speculative and risky than average, Janelia provides a supportive environment."
Janelia seemed like a natural fit for another reason. The campus was modeled in part after AT&T's Bell Laboratories in Murray Hill, New Jersey, and Jain had "grown up" scientifically under the supervision of former Bell Labs researchers. "I am fond of saying that I have been passed on from one Bell Labs person to another," says Jain. Jain's undergraduate research advisor Saul had worked at Bell Labs where he met Seung, Jain's supervisor at MIT. At MIT, Jain collaborated with Winfried Denk, director of biomedical optics at the Max Planck Institute for Medical Research and a senior fellow at Janelia. Denk had also worked at Bell Labs. "Coming to Janelia Farm," says Jain, "feels like I am maintaining a certain tradition."
At Janelia, Jain plans to further improve the accuracy of the algorithms he developed as a graduate student. In addition, he will design new tools that will allow people to work with the results of computer analyses in more efficient ways. "The computer does the work but a person still has to go through to make sure it is accurate. We are designing ways to focus human attention on those parts of the data that are more likely to be inaccurate," explains Jain.
He also plans to start applying his computational tools to understanding how the brain functions. To this end, he is establishing research collaborations with neuroscientists at Janelia and elsewhere who can use the techniques he has developed to analyze specific local circuitry in selected parts of the fly brain. "One of the nice things about being at Janelia is that I don't have to go far to find collaborators," says Jain. "There are plenty right here."