Life in the computer

Andrey Palyanov (Institute of Informatics Systems SB RAS) about artificial lifeAlla Arshinova, Computerra
Creating a digital analogue of a living organism is difficult for two reasons.

Firstly, this is not an easy task purely technically. Secondly, there are almost no creatures in whose nervous system biologists have figured out to the extent that it is possible to talk about any models. Novosibirsk scientists have managed to build a virtual nematode double (C. elegans) – one of the few animals whose nervous system has been thoroughly studied. Andrey Palyanov, a researcher at the Institute of Computer Science Systems named after A. P. Ershov, SB RAS, who led the modeling of C. elegans, told Computerra about this project.

– Andrey, tell us, what does neurocybernetics do? – Neurocybernetics is a scientific field that studies the principles of operation and patterns of control processes in living nervous systems at all levels of the hierarchy.

Neurocybernetics studies both individual neurons and complex functional modules, and the most difficult task facing this science is the study of the brain as a whole (it's no secret that today there are more questions than answers).

The main method of neurocybernetics is mathematical modeling, while data from a physiological experiment are used as the source material for creating models. One of the most promising areas of neurocybernetics is modeling based on neural networks.

Here we mean models of biological neural networks, the purpose of which is to reproduce the properties of real biological neurons, starting from the basic ones and ending with increasingly small details and features. This may be taking into account the three-dimensional structure of neurons and interneuronal connections, taking into account the types of synapses (interneuronal connections), taking into account the type of neurotransmitter used, changing the parameters of a frequently used neuron, as well as, possibly, implementing a mechanism for the formation of new connections between existing neurons and embedding new neurons into an existing system.

– Why did you decide to do this job? – Even when I was studying at NSU, I was interested in tasks related to modeling living systems or their individual components.

That's why I chose the Department of Chemical and Biological Physics at our physics faculty. The thesis was related to modeling the structure of RNA, the candidate's thesis was related to the problem of modeling the mechanisms of laying protein molecules. At the same time, personal interest in the principles of the mind in general and artificial intelligence in particular grew.

Despite the impressive development of science and technology, there is still no artificial computer system with intelligence, much less consciousness. Attempts to create artificial intelligence without delving deeply into the biological details of the structure and functioning of the nervous system of living organisms have not led to success. Yes, they have become one of the essential reasons for the rapid development of computers, the Internet and digital technology in general, but the task itself, in essence, has remained unresolved.

Nature in the course of evolution, as a rule, creates extremely optimal solutions. The human mind is the product of millions of years of evolution, so it is rather presumptuous to expect to reproduce it by simply grasping the basic principle and limiting yourself to many thousands of times fewer structural elements and computing power.

A much more promising way is to create the most accurate functioning computer copy of a living nervous system. The study of the nervous system of the simplest possible creature can be the first step, which presumably will mark the beginning of a new direction, will allow you to work out the technology and understand all the details. That's what we're doing.

Our research group, although still small, includes specialists in the field of mathematical modeling, programming, biophysics, molecular biology and neurobiology. A specialist in the field of neurophysiology, working in the framework of the main direction with real cultures of living nerve cells, also participates in the work as a scientific consultant.

– What is the purpose of the project? – The goal is to create the world's first virtual organism controlled by an electronic copy of its biological neural network.

Using the example of this simple organism, we want to find out whether a copy of the nervous system built on the basis of an organism's connectome (a collection of data describing all its neurons and interneuronal connections) gives the same behavior in a number of test situations as the original. How close to reality is our understanding of the principles according to which neurons and neural systems function, is there no need to significantly revise the concept?

– Why the nematode C. elegans? – By and large, we had no choice. C. elegans is the only organism today for which the entire or almost the entire connectome is known – a set of neurons, interneuronal and neuromuscular connections, sensor cells and a number of parameters describing these systems.

There are simply no other organisms so studied in this regard. Nobel Laureate in 2002 J. Salston, who received the award just for his work in this field, not for nothing said about him: "When we solve the worm, we will understand life."

As for the biological characteristics of the object, they are also unique in many ways. C. elegans is a free–living soil nematode, a small worm about a millimeter long.  The short life cycle, the period of growing up, calculated in several days, made it extremely convenient for research in the field of genetics.

In 1998, the genome of C. elegans was sequenced. Neuroscientists also paid well-deserved attention to him. The beginning of a large-scale study of the nervous system of C. elegans can be considered the work "The pharynx of Caenorhabditis elegans", Albertson & Thomson, 1976, devoted to the study of the cerebral nerve node of C. elegans and "The Structure of the Nervous System of the Nematode Caenorhabditis elegans", White et. al., 1986, the result of which was the receipt of experimental data about most neurons and interneuronal connections.

As it turned out, the nervous system of all individuals of the same sex is identical: 302 neurons, about seven thousand interneuronal connections, 95 muscle cells and several dozen sensory cells of different types. Another significant advantage of this organism in terms of modeling is transparency in the optical range.

– Why is transparency a virtue? – More complex organisms tend to reliably protect their central nervous system.

In all vertebrates, the brain is hidden inside the skull, and insects and crustaceans have acquired a strong external framework. All this makes it impossible to observe their brains, especially under a microscope, directly in a living organism. But C. elegans – please. There are hundreds of micrographs of individual neurons, their processes, muscle cells, and so on.

– How does the nervous system of a nematode work? – Until 1990, no one seriously considered the ability of C. elegans to plasticity of behavior and the use of experience for learning, but then, as a result of a number of studies, the opinion of scientists on this matter has changed significantly.

Since the first work devoted to learning and memory in C. elegans, it has been shown that the nervous system of this nematode is not as unchanged during the life of the organism as it initially seemed. On the contrary, it turned out that it was designed by nature as a miniature, but at the same time perfectly honed mechanism for extracting experience, memorizing and learning based on signals from mechano-, chemo– and thermoreceptors, through which the nematode perceives the environment.

Recently, it was also discovered that despite the absence of eyes, C. elegans responds to changes in light, and scientists have even discovered several neurons associated with processing these signals. The nematode can be trained: to approach or, conversely, to avoid sources of taste, smell or temperature changes, which, based on previous experience, make it possible to predict the presence or absence of food. The worm also exhibits associative forms of learning, such as the development of classical and differentiated conditioned reflex, and has abilities for short-term and long-term memory (Catharine H. Rankin, Current Biology, 2004).

According to current ideas, it is believed that the nervous system of any adult individual of one of the sexes invariably contains 302 neurons and the formation of new ones, even as a result of training, does not occur. However, these observations do not exclude the formation of new interneuronal connections and reconfiguration of the parameters of existing ones. Exactly how this happens remains to be seen, both through experimental work and with the help of computer modeling.

The data on the structure of the nervous system used in the simulator correspond to an adult that has been digitized, and must contain all the basic behavioral programs according to which the virtual C. elegans should behave plausibly. However, the learning abilities described above will "work" only if it is possible to identify the mechanisms behind them and supplement the model with them.

– Is it enough to simulate only the nervous system? – Creating a model of a "nervous system in a vacuum" is not at all as interesting and fruitful as simulating simultaneously a set of interconnected systems – nervous, muscular and sensory, based on a flexible body frame placed in a virtual environment – a physical simulator.

This will allow the nervous system to receive realistic sensory input from the environment, changing in response to the actions of the worm itself, which will be carried out as a result of the activity of the muscular system, controlled in turn by the nervous system. And with the help of the 3D visualization module, researchers can observe both the neural and muscle activity itself in the smallest detail, and its result - the behavior of a virtual nematode.

Many serious laboratories in the USA, Europe, and Japan are working on the research and modeling of various C. elegans systems. However, this is exactly how we have set the task for the first time. We tried to combine all the existing data into a single multifunctional software package. The successful implementation of the project will allow us to gain new knowledge about the mechanisms of the nervous system as a whole and to study in detail, at the level of individual neurons, the principles of the organization of functional blocks of biological neural networks, which in the future can be used to design artificial neural networks of a new generation.

– What computing power and software do you use for modeling? – So far, a powerful personal computer is enough for the simulator to work properly, although it is clear that as the system is supplemented with large amounts of new data, this balance may be disrupted.

If the performance is not enough, there are ways to optimize and parallelize.

As for the software, the arsenal is quite simple. The simulator itself, both physical and neural, is developed independently and implemented in C++ using the standard STL library, all 3D visualization is implemented using OpenGL technology, which allows you to run a software package under a particular operating system with minimal code adaptation.

– What does the nematode you created "know"? – We have proposed an original scheme of the flexible skeleton of the nematode body, repeating the shape of the biological prototype, optimized for attaching muscle cells in the same positions in which they are located in a real worm.

We were the first to propose a 3D model of the C. elegans muscle system, in which each muscle cell (out of 95) of a real organism will have its own analogue.

Geometric and mechanical properties are reproduced as accurately as possible, including the positions of neurons and the architecture of connections between them.

There are a number of significant reasons for this. Due to the peculiarities of the structure of C. elegans neurons, for their modeling, it is necessary to take into account in the calculations the real pathways of connections between neurons, the change in the signal along them and the time of its propagation. Our concept meets these requirements and provides an ideal way to visualize the structure of interneuronal connections, including nonlinear sections and branching regions, as well as display the dynamics of neural activity.

To do this, the linear connection between each pair of neurons will be replaced by a series of consecutive segments set by a system of intermediate points that will repeat the course of the real connection. Each such point will simultaneously act as a "transfer station" for calculating signal attenuation and providing the necessary time delay, and also in the appropriate situation can become a branching point.

The data for entering this kind of information into the model is mainly extracted from micrographs of neurons, which is a time-consuming process. We managed to simplify one of the stages within this task as much as possible by creating a visual 3D neural network editor.

As a result, we have so far managed to "launch" only about 10-15% of the entire nervous system, mainly related to the ventral nerve cord (abdominal nerve chain), which controls the muscular system and provides basic motor activity (sinusoidal forward or backward movement).

We can already observe a realistic forward movement, its change to the opposite movement when touching an obstacle (so far by artificially switching the phase of the sinusoidal signal applied to the muscles).

Our nematode also "knows" how to turn 90° and continue moving along the obstacle. More complex movements and behavioral patterns (changes in speed or direction, turns, searching behavior, stimulus avoidance reaction, etc.) are achieved with the participation of additional signals from the brain nerve ganglion, which is still far from fully functioning within the framework of the model.

The possibility of implementing a sensor system is embedded in the simulator and planned, but so far this is one of the most difficult parts of the task, since the encoding of signals from receptors is practically unknown. Detailed consultations with specialists studying the nervous activity of this nematode experimentally are needed. We are also working on establishing contacts and cooperation with research groups working in this area.

– What will be the next step? – Despite the serious groundwork, even for a complete simulation of the nervous system of C. elegans, a lot of work will still be required.

First of all, it is necessary to develop a methodology for modeling biological neural networks and refine and complicate the models of the neuron and interneuronal connections and interactions. This will happen as the project progresses and new information is obtained, including as a result of collaboration with colleagues studying the nervous system of C. elegans experimentally. In addition, we plan to improve the software tools to ensure high efficiency of working with the modeling environment.

The next step is to consistently configure, study and "debug" individual fragments of the neural network, including on the basis of published information about the study or modeling of these fragments, which previously could not be verified in practice due to the lack of a functioning model of the organism.

Despite the fact that the architecture of the C. elegans neural network is known, many mechanisms of its operation have not yet been explained. For some fragments of the nervous system, there are supposed explanations of their mechanisms of action and theoretical models, and for many there are none at all. All of them require verification, which can be carried out just with the help of a multifunctional interactive modeling environment, the creation and use of which our project is dedicated to.

If we can do all this, then we plan to introduce known data concerning the sensory system into the model and connect it to the nervous system.

– What other interesting studies are being conducted in this area?– One of the most ambitious projects is The Blue Brain Project, launched in 2005 with the simulation of a fragment of the neocortex (a new cerebral cortex responsible for higher nervous functions) of a rat, based on the results of 3D digitization of 10,000 neurons and 3•10 7 synapses of real nervous tissue.

It took fifteen years of painstaking experimental work to accumulate this information. The researchers successfully completed this phase and moved on to the next one – modeling a fragment of the human neocortex. This is a very bold, ambitious project, however, much is still unclear here. For example, not everything is clear with the input and output information coming into this fragment, the role of neocortex connections with other parts of the brain, which are not yet present in the simulation, is unclear. It is also unclear how to understand in such conditions whether this fragment works correctly or not.

The currently existing human brain fragment in the form of a model is equivalent to about 1/10000000 of a part of the whole brain. Thus, here we see an attempt to approach the problem of studying the principles of the nervous system from the other end – instead of modeling the simplest being, on the contrary, take up the most complex existing nervous system, but simulate a small fragment of it.

Among the most important achievements, it is worth noting the technology developed in 2007 at the Massachusetts Institute of Technology for digitizing the structure of nervous tissue with high resolution (MIT Technology Review).

There is another approach to the same problem, developed in 2010, called Array Tomography. According to the authors, this is "a new high-performance method that provides unprecedented opportunities for visualizing the molecular architecture of living tissue at high resolution." It includes the following main stages: automatic execution of ultrathin tissue sections; creation of arrays of sections lying sequentially in the same plane; staining of preparations and obtaining their images; computer reconstruction of a three-dimensional image and then; spatial (volumetric) analysis of the resulting images.

A recent paper published in October 2010 in the journal Nature is also of considerable interest. Scientists from the Salk Institute for Biological Studies, San Diego, California have developed a system that allows you to simultaneously monitor the neural activity of the output layer of the retina through several hundred channels implemented in the form of a matrix. This matrix provides both a very high spatial resolution, comparable to the size of a single neuron, and a time resolution that allows for more than ten million records per second.

By stimulating the signal entry point and high-speed reading of the exit point, it was possible to determine the connection scheme of cells and the entire structure of the neural network of the eye that forms the visual perception of the world. This allowed the researchers to recreate a complete picture of information processing on the way from light hitting the cells of the retina of the eye to its arrival to the fibers of the optic nerve leading to the brain.

– And if you think globally, what are the results of the research that you are doing? – As a result of the work on the project, it is planned to study, analyze and reproduce the neural mechanisms of C. elegans, only a small part of which has been understood and explained to date, to find out whether it is possible on the basis of modern models of a biological neuron to obtain the behavior of a "virtual" organism close to the original, and thereby lay the foundation for modeling and studying the nervous system significantly more complex creatures.

Probably, as a result of studying and analyzing the principles of the nervous system at different hierarchical levels, it will be possible to identify stable functional neural patterns, which in the future can be used to design artificial neural networks of a new generation based on biological prototypes. Thus, the project may be of significant interest for the fields of science related to neuroscience, cybernetics and artificial intelligence.

Even more global reflections can only be of the nature of a forecast and belong, rather, to the field of science fiction. The central goal of research in the field we are considering is to develop a scientific and technological base sufficient to create a functioning digital copy of the human brain, functioning and having consciousness, like the original.

Further prospects are more difficult to assess, however, if we accept the possibility of the existence of a person's personality in digital form, then this automatically leads to the possibility of unlimited life expectancy, solving the problem of "teleportation", which in this case will be reduced to copying the "electronic brain" over the network to the desired point of the planet (which means that the problem of traffic jams along with the threat is being solved at the same time depletion of fuel reserves), and the possibility of unlimited increase in intellectual capabilities through man-made modification of the architecture of their own brain by its most gifted owners or specialized firms.

Portal "Eternal youth" http://vechnayamolodost.ru
27.01.2011