EyeWire - keep learning brain secrets

Mural concept for the city of Cambridge, Massachusetts. The author - Daniela Gamba.

The EyeWire game, about which I have already written an article here, continues to turn out new material. Scientists in the laboratory of Sebastian Seung do not rely only on the game and only on the efforts of science volunteers to promote research in the field of brain research, but it certainly has scientific value.

According to official statistics of the game for 5 years, 265,000 people from 186 countries have registered (data are based on player profiles and may be inaccurate). Analyzed 10 million "cubes" with the processes of cells, completed models for approximately 3000 cells. Is it a lot or a little? It took the team of scientists more than 10 years to fully describe 302 nematode nematodes (C. elegans) and the connections between them, the work was completed in 1984. The article states that, although computer help was used, scientists did most of the work on 8000 images manually. Progress does not stand still and today we can “entrust” the process of reconstructing neural networks with computer neural networks, but first these networks need to be taught. The problems of this process are described , for example, by a different team of scientists doing a reconstruction of the Drosophila neurons. On the other hand, it is impossible to speak of a complete description of neural connections (connections) in more complex organisms without some radical improvement in the speed of modeling. In small animals and birds, the nervous system consists of 10 to 500 million neurons, in large ones more than a billion, in particular, about 86 billion neurons in humans.

Despite the fact that attempts to fully describe the connectoms of whole animals are still far from success, even individual discoveries greatly enrich the rapidly developing science. For example, a publication on the finding of new compounds in a well-studied part of the Drosophila brain as a result of the reconstruction of its connectom. Work is also underway on derivative studies: for example, having a model of the nematode nervous system, build a general computer model of the worm itself, simulating its behavior. The OpenWorm project is engaged in it - article on Habré, news on Habré.
')
But back to the eyewire. At the same time as scientists work with raw data from an electron microscope, volunteers in EyeWire actually create a set of validated data. A side effect of creating such a set of reference data is the discovery of new cell types and subtypes, which scientists have not previously encountered. Human solutions also help in cases where it is impossible to distinguish the course of the branches of a neuron at a section due to their intersection with another neuron. A simple connection according to certain characteristics, such as brightness, does not make it possible to understand that we have a branching of the neuron under study or part of another branch ( examples on the game site). In addition to the illustrations, from which it becomes clear why it is difficult even for a person to reconstruct cell processes, I prepared a video with the game process. The video recorded the reconstruction of several "cubes" with the processes of the cell, and at 1: 40-2: 20 one of the defects is visible (non-contrast intersection) and the extension of the process is behind it.

The data with which EyeWire is currently working is divided into 2 large parts: neurons from the mouse retina and more complex cells from the brain of the zebrafish. The first dataset has a size of 16 sectors, each 50x50 microns, the number of cells is estimated at 2000. Work on the dataset started in January 2016, as of December 2017 (5th anniversary of the game), the last cells from sector 9 were investigated, by May 2019 - mid 14 sector. At the same time, we do not take glial cells for construction - they are auxiliary and (as far as I know) do not form synapses with other neurons. In the laboratory, there are those who help with the organization of the process (game makers) and those who work with already processed cells and their connections, considering them from the point of view of modern views on the functionality and connections of this part of the brain.

In the last article I mentioned Zoouniverse as an example of a platform where quite scientific tasks of various directions are solved with the help of amateurs. Eyewire has a number of differences from the Zoouniverse scenarios, which required writing its own engine and why the community in the game is not necessarily related to other projects.

Simulation takes place in three dimensions, the picture of the neuron is completely voluminous.



The process begins with the fact that the game maker finds a previously not reconstructed cell nucleus in dataset and removes the first processes from it. Then, the game maker gives a link to this “germ” and asks experienced players to evaluate the complexity of the cell (1 or 2). At some point, the cell is shared, from that moment on, any member of the game who has passed the training material of the appropriate level can work on it. In the lighter cells of type 1, where novices have access, the cube is considered complete with three reconstructions, and if at least one of the reconstruction segments abuts the wall of the cube, then the game searches for a neighboring section in dataset and substitutes it for further play. In case of a conflict of opinions, the game will reckon with the majority. To correct the opinion of the majority can be players with access to editing (the possibility given by the administration with experience of several months and high accuracy parameters of previous reconstructions) and the administration itself (game makers). Also, players with access to editing place the label “completed” on the cube, and it drops out of the gameplay. As soon as all the cubes of the cage are labeled "completed" from two different players, game makers take the cage for the final check. In the laboratory, they have a visualization of all reconstructed cells and the connections between them, which allows us to find flaws and problem areas.

Despite the relatively small number of active participants (100-200 people give about 90% of the contribution per month), it is already clear that on this date both the game and the study of the brain of small animals with its help will not end. For experienced players, the danaco dério brain is also available in parallel, and a team of game makers and developers are planning a new project called Neo, in which axons will be studied (and not dendrites, as now). I did not like the preview of the game, which I saw (it felt like they were guided by children of middle school age), but I would look forward to what would come of it. Deadlines, as is often the case, have been postponed more than once.