Seung and Denk aim to dramatically speed up the tracing process, which takes a single graduate student weeks to complete, with automated machine-learning algorithms. The researchers use data from a manually generated wiring diagram to train an artificial neural network to emulate the human tracing process. They can then use the resulting algorithm to analyze new chunks of brain tissue. To date, they’ve been able to speed the process about one hundred- to one thousand-fold.
The researchers presented their initial findings to an awed crowd at the Society for Neurosciences meeting in San Diego earlier this month. They showed the three-dimensional reconstruction of part of the rabbit retina called the inner plexiform layer, which is a piece of neural tissue at the back of the eye that senses light and sends visual information to the brain. (See a movie of the reconstruction here.) “But we need to improve 106-fold or more,” says Denk, who estimates that this would shrink the three billion person years it would take to trace a cortical column down to about two years. “I’m confident in the end that we will be able to do it,” he says. “But I don’t know how long it will take us–if we’re lucky, maybe a year or so.”
Earlier this month, scientists at Harvard described a new method of tracing neurons in the living brain by labeling them with up to a hundred different colors. (See “The Technicolor Brain.”) “We’re starting to think about wiring diagrams as being fundamental,” says Jeff Lichtman, one of the researchers who developed the technique.
Researchers say that the two approaches will likely be complementary, allowing scientists to look at neural circuits of different dimensions. Eventually, Seung aims to generate maps of the complete fly connectome, as well as partial wiring diagrams of interesting locations in larger brains, such as the hippocampus, olfactory bulb, and retina.
Just exactly how much light these maps will shed on the brain is still somewhat controversial. “Just knowing the [wiring] data won’t take us far if we don’t put it in the framework of processing and transferring data in the brain,” says David van Essen, a neuroscientist at Washington University, in St. Louis, and president of the Society for Neurosciences. Seung and others eventually hope to generate maps that incorporate the biochemical and physiological properties of various cells into the wiring diagrams.
Hear more from MIT at EmTech MIT.