- Special Sections
- Public Notices
The tiny subject, nearly impossible to swat when loose, was stuck in wax inside a plastic tube with its head sticking out. Through a small hole in the back of its head, a microelectrode connected to an electronic amplifying system tapped into a central neuron that was funneling information from thousands of other neurons. Finding that one microscopic neuron was not the hard part. “The neuron doesn’t fire unless you have horizontal motion,” said physicist Ilya Nemenman, of Los Alamos National Laboratory. “After you connect the electrode to the amplifier and the speaker, you start inserting it with some levers into the fly brain while moving your hand from left to right. When you come close (to the target neuron) you can really hear it.”Then, too, the fly was subjected to a shaking and spinning environment that simulated acrobatic maneuvers, and the experiment took place outdoors, in a way that recreated the kinds of changes a fly might see evading a predator or chasing a companion. In an intricate collaboration between theorists and experimentalists, a team of researchers has begun to crack the neural code that a blowfly uses to communicate sight.In the process they have broken new linguistic and computational ground and established in principle the uncompromising efficiency of the natural world.Nemenman with partners Geoffrey Lewen and William Bialek of Princeton University and Robert de Ruyter van Steveninck of Indiana University, published an article in the Public Library of Science (PLOS) Computational Biology Journal Monday, describing their peer-reviewed findings on “Data Processing Through a Fly’s Eye.”The work has taken nine years, Nemenman said Monday in a telephone interview, much of it trying to figure out basic dimensions of the neural network, how much bandwidth there was, what its channel capacity might hold and how much signal information was contained in what amounted to a sequence of spiked pulses with varying intervals of silence.A few more years perfecting the right tools for the job, and the effort has begun to give results.These coded signals, the researchers found, are not only more precise than had previously been understood, but their precise timing suggests an even more elaborate message.“We find, that significant amounts of information are represented by details of the spike train at millisecond and sub-millisecond precision,” they wrote. A millisecond is one-thousandth of a second, a scale that is also used for access times on computer hard drives. For example, Nemenman said, a code word in fly-speak might be a silence, followed be two spikes, and after a short but measurable pause, another spike.“The message is telling the fly more than ‘go left’ or ‘go right,’ and more like the whole story up to now,” Nemenman said. “You went right and then left.”This interest by the fly in its own recent trajectory was one of the surprises of the current research. “You might think the only information might be where the fly is going, at what precise angle, but the fly is more intelligent than that,” Nemenman said. The fly also needs to know the context of its present maneuver.An important issue was how much redundancy there might be in the message. Condensing a typical computer file by a zip program, for example, weeds out the junk and compresses the information to its essential ingredients.An algorithm was needed and found to describe what Nemenman calls the “entropy” of the system, its variability, in order to compress the amount of data needed to a realistic level, because the blowfly doesn’t live long enough. “You record from the fly with an electrode in its brain, riding a rotating machine and shaking wildly, maybe two hours a day for two or three days,” Nemenman said.Without the quantitative analysis made possible by new mathematical methods, merely capturing the data would take 200 days. “Recording from a fly for so long is impossible,” he said.The team also shared a fundamental respect for the highly perfected sight-flight coordination of the blowfly.“As it moves around, the blowfly is subjected to an amazingly complicated jumble of visual information from a complicated world,” said de Ruyter van Steveninck, a biophysics professor who has been working with the blowfly and its visual acuity for years. In a press release from Indiana University, he added, “But because of its agility in flight, it is a reasonable guess that the way in which it analyzes and encodes motion information is nearly optimal.”“A theory thus could start by taking these examples of near optimal performance seriously and articulating optimization principles from which essential aspects of neural circuitry could be derived,” wrote Bialek, one of the theorists in the group in the abstract of a talk at lab sponsored conference, “Grand Challenges of Neural Computing.” The conference took place in February 2007 in Santa Fe.There are no immediate applications besides other potential uses for the algorithms, Nemenman said. But the discovery that biology doesn’t just count the spikes, but the time between them, may also have implications for neural science in general and for artificial systems of computation as well.The team is now designing experiments to analyze the neural code in more detail.