MOSCOW, November 5, Tatyana Pichugina. In 2023, the European Human Brain Project to create a model of the human brain ends without reaching its goal. The chosen path turned out to be too difficult. Russian scientists proposed solving the inverse problem – looking for parameters under which the model reproduces experimental data. Candidate of Biological Sciences Ivan Mysin, senior researcher at the Laboratory of Systemic Organization of Neurons named after O. S. Vinogradova at the Institute of Theoretical and Experimental Biophysics of the Russian Academy of Sciences, spoke about how this approach works to describe brain rhythms.
Scientists have thought about the brain model
In 2013, the EU launched a large-scale Human Brain Project with funding of a billion euros. Its founder, Henry Markram from the Federal Institute of Technology in Lausanne, believed that in ten years it was possible to create a model of the brain at the cellular level. The idea initially aroused skepticism in the scientific community, writes Nature.
Markram led the large Blue Brain project on computer modeling of the visual cortex of the brain. If you have a lot of experimental data and computing resources, the goal is achievable, he argued. They believed the scientist, ten years have passed, but the model has not yet been created.
When starting to work in this direction, Ivan Mysin often heard two opposing opinions from senior colleagues. Some believed that it was necessary to model individual empirical phenomena, others – the brain as a whole. Later, the scientist realized that these were not different points of view, but a manifestation of the crisis: no one really knew how to create these models.
Modeling makes it possible to explain experimental data and find unexpected connections in them. In brain science, this is one of the main research methods.
This is important for understanding the functioning of the brain and solving applied problems, primarily the treatment of Alzheimer's disease, epilepsy, and parkinsonism.
The problem with large models is that they require many parameters that are not easy to measure. One approach is to solve small problems that explain individual experimentally discovered patterns. A small model requires a small number of parameters that can be adjusted based on a hypothesis about the mechanisms of the phenomenon being described. Ivan Mysin devoted one of his first works to how the theta rhythm is formed and transmitted to the hippocampus. The model describes the interaction of only six populations of neurons.
To model the entire brain, as envisioned by the Human Brain Project, it is necessary to have data on large populations of neurons. It was supposed to experimentally measure all these parameters, put them into the model and see what happens. However, this approach did not work for several reasons. Firstly, the data are heterogeneous – obtained on many animals, they are not accurate enough and are not always comparable. Secondly, large computing power is needed. For Blue Brain, they took the most powerful supercomputer at that time used for non-military purposes – Blue Gene. However, no breakthrough results were achieved. In such large-scale projects, they try to build too detailed models, which is often redundant. Many processes can be described more simply.
Hippocampus and its rhythms
The most studied brain structure is the hippocampus. It is not without reason that this is a model object for neurobiology. Essentially, this is a simplified version of the cerebral cortex, evolutionarily more ancient. It is important for the formation of attention and memory.
The hippocampus has several zones that are functionally connected to each other by bundles of fibers that are visible under a microscope. This helped to reveal many of the secrets of memory back in the 1970s. One of the most significant achievements belongs to the Soviet neurobiologist Olga Sergeevna Vinogradova, who worked in Pushchino. She experimentally established that high-frequency stimulation of a bundle of fibers leads to an increase in responses, and low-frequency stimulation, on the contrary, leads to a decrease. This was one of the first evidence that brain plasticity is achieved by changing the strength of connections between neurons. Today, the concept that memories are stored in connections between neurons is generally accepted. The experiment on stimulating fiber bundles in the hippocampus has become a classic.
Throughout her scientific career, O. S. Vinogradova studied the role of the hippocampus in memory formation. She led experiments to record neural activity in all areas of this part of the brain. This helped develop a theory known as Vinogradova's bicycle, in which the hippocampus “decides” what information is new and worth remembering and what should be ignored. This is what attention is based on.
Then, in the 1970s, other scientific groups discovered neurons that determine the position of an animal in space, thereby showing that the hippocampus is a repository of spatial memory. Later they found nerve cells that encode the stages of behavioral acts in animals, and in people that react to images of famous actresses and politicians. This meant that the hippocampus stores all types of memory as a sequence of images.
Nowadays, many such empirical patterns about the encoding of information in the hippocampus have been established. The challenge is to summarize them into a coherent theory that can be implemented in a computer model.
Hippocampal neurons are 90 percent pyramidal cells. They are the ones who encode information. The rest are inhibitory neurons that regulate the activity of pyramidal cells, that is, the generation of impulses (or, in other words, action potentials). They can be recorded by instruments and thus monitor the transmission of signals between neurons. Almost all of their activity is rhythmic.
Theta and delta rhythms are expressed in the hippocampus. The first has a frequency of four to eight vibrations per second. The theta rhythm is recorded when an animal is studying something, searching for something, or reacting to an external stimulus, for example, a sharp sound. This is the rhythm of attention. It is expressed when you need to remember information, remember, compare with what is expected.
The delta rhythm (frequency 1-4 hertz) appears in a state of rest and sleep. It is necessary for consolidating memory.
The theta rhythm was discovered in the 1930s and for a long time they could not understand how it arises. It turned out that there is a special structure of the brain – the medial septal region, whose signals cause hippocampal neurons to work rhythmically.
According to one hypothesis, rhythms organize the activity of neurons and the flow of information. When brain cells are activated together, the connections between them are strengthened.
At the behavioral level, this leads to the formation of associations. As in Pavlov's experiments: the light bulb lights up – saliva is released. Some neurons were activated, followed by others. A stable association has been created.
The work of the brain is characterized by orderliness – first one group of neurons is activated, then another. To strengthen the connection, pulses are needed with a difference of no more than 30 milliseconds. Rhythms increase the likelihood of coincidence in the functioning of neurons and the emergence of associations. This probably speeds up learning.
1 of 3
2 of 3
3 of 3
1 of 3
2 of 3
3 of 3
The model predicts a new property
Rhythms are formed by tens of thousands of neurons. In a sense, this is their average activity. Modern techniques make it possible to record the activity of hundreds of individual neurons. Using this data, you can build a model that predicts the behavior of all neurons.
Scientists from Pushchino collected experimental data from the open press on inhibitory neurons of the CA1 field, the most studied area of the hippocampus. They work in a certain order, for example, first one type of neuron, then another, then the first again, and so on. There are inhibitory connections between neurons that organize this order. But how? A hypothesis has arisen that short-term plasticity is involved in this – a change in the strength of connections between synapses, the main mechanism for the formation of short-term memory.
There are 18 known types of inhibitory neurons. The scientists selected just seven and simulated the network. However, even this incomplete model required tuning 301 parameters for the equations. Then they developed an original mathematical approach, which made it possible to reduce the solution of the problem to what is used in AI – for example, in ChatGPT.
Scientists intend to take on more complex projects that require setting up thousands of parameters. This will make it possible to build large-scale models of neural networks in the brain. It is possible that eventually it will be possible to make a model of the entire brain. Maybe not as fast as we would like.