Combinations

The effort to decipher the way in which the nerve cells in the brain talk to each other is similar to learning a new language by listening only to the conversations of the speakers of the language. At first this may seem an impossible task to us, but over time we will begin to pick up and understand basic words and phrases that repeat themselves

Dr. Elad Schneidman from the Department of Neurobiology at the institute simulates the effort to decipher the way in which the nerve cells in the brain talk to each other for learning a new language by listening to the conversations of the language speakers only. At first this may seem an impossible task to us, but over time we will begin to pick up and understand basic words and phrases that repeat themselves. When we reach a situation where we already understand about a thousand or two thousand words, we will also have a preliminary understanding of grammar, and thus we will be able to incorporate new words into our conversations.

Most of the information we have about neural communication in the brain comes from studies that deal with the "letters" and "words" that make it up - that is, the activity of individual nerve cells, or small groups of cells. This information is obtained in experiments in which the "firing" of electrical signals, which are fired by single nerve cells, or pairs of cells, is measured. These attempts are somewhat similar to trying to understand a book by reading a small number of words from it. In Dr. Schneidman's opinion, the really interesting conversations take place between larger groups of nerve cells. His research attempts to decipher the basic rules of communication between nerve cells and their group behavior. To this end, he observes patterns of electrical activity in networks of about 100 nerve cells, and tries to understand the interrelationships between them.

Very few researchers attempt to accurately study such large groups of cells. The difficulty also stems from the fact that 100 neurons present a huge abundance of possible activity patterns - on the order of 1030 patterns. Therefore, any attempt to extract useful information from such a network seems to be an impossible task.

Dr. Shneidman, together with the research student from his group, Elad Ganmor, and Dr. Ronan Segev from Ben Gurion University, approached this challenge equipped with a combination of experimental tools and mathematical models. For the experimental part, the scientists took pieces of retina from the eyes of salamanders and archer fish. Each piece of tissue, about two millimeters long and wide, contained hundreds of nerve cells, of which the electrical activity of 100 cells was recorded for hours. The researchers projected nature films onto these pieces of retina, and examined the electrical signals that the nerve cells send. "The electrical activity of the nerve cells in the retina is actually the 'output', meaning the result of the 'calculation' performed by the retina on the visual input, which is then sent from it to the brain," says Dr. Schneidman. "The cells of the retina and the cells of the brain are found on one neural circuit, and the communication in the retina is the same as the communication between the brain cells. When this neural network is exposed to different scenes, we can see unique patterns of activity. It is interesting to note that we were able to distinguish activity patterns that obey a unique 'grammar', which appears, apparently, only in response to natural scenes, but not in response to scenes of 'white noise', or to the projection of unnatural characters".

To reveal some of the basic rules of nerve cell activity, the scientists used a mathematical model similar to a model from the field of physics, which was developed to study the behavior of a large number of magnets in magnetic fields. A similar model is used both in statistics and to study how machines learn. In all these fields, the complex behavior is created as a result of interrelationships between a pair of factors: attraction and repulsion in the case of magnets, "on" and "off" states of binary variables, electrical firing and silence of nerve cells. When the scientists entered the data collected in small networks into the model, a good match was obtained between the model and the experimental findings. In larger networks a fairly good fit was obtained, except for a few points that did not fit the model. Upon closer inspection, the scientists realized that these points belong to the most common activity patterns, which present a more complex grammar. More precisely, they expressed a mismatch between nerve cells that cannot be explained by relations between pairs of cells alone. As a result, the model they developed identified the rare combinations well, but was less accurate for more common combinations. Similar to a person learning a foreign language, who needs to learn to say "I want to eat" before proceeding to order a four-course meal, Dr. Schneidman and his colleagues also realized that a full understanding of the language of the brain requires dealing with "everyday" expressions as well. The challenge they faced, therefore, was to find one model that would "cover" both the common and the rare patterns.

To their surprise, the scientists discovered that a seemingly negligible change in the way the activity of the cells is represented in the mathematical model gives a simple and effective way to deduce the grammatical rules of communication between nerve cells: instead of marking a quiet nerve cell with (1-) and an active one with '1', as they were marked In the original physical model, the negative and positive poles of a magnet, they used '0' and '1'. This change - which is apparently purely computational - had a huge effect on the arrangement of the organs in the formula. They realized that this change would only occur in one particular case: the case where the neurons in the network are relatively rarely active. This is exactly the situation in the brain: most of the time, most of the neurons in the brain are inactive. The small change in the model made it possible to translate the common combinations as well, thus revealing the basic interrelationships between nerve cells. In addition, the examination of the differences between the various combinations allowed the researchers to identify different rules than those expressed in the common patterns.

In fact, from an almost unimaginably complex network of possible interactions, the scientists were able to obtain an incredibly accurate picture of the nature of communication in a large group of nerve cells. "We were able to put together a basic 'grammar book' that presents millions of patterns of electrical activity, which are created from about 500 common combinations based on connections between pairs, triplets and quadruplets of nerve cells," says Dr. Schneidman. "It seems that it is possible to learn the grammar of the language of brain cells, assuming, of course, that we know which examples to choose." He believes that it is possible to learn the language of nerve cells because it is structured somewhat like the natural languages ​​we know, and the reason is that one part of the brain needs to learn the language of another part. Thus, for example, it is possible that the constant repetition of common combinations is the way in which the nerve cells acquire their communication abilities, and continue to understand each other.

Following the new insight into the nature of intercellular communication, the researchers were able to decode the visual information received from large groups of retinal cells. Dr. Shneidman believes that with the new approach it will be possible, in the future, to get a detailed picture of the activity of large groups of nerve cells in different parts of the brain, and even "read" the information encoded in these networks. This ability may pave the way for new approaches to the treatment of various neurological problems.