Shulou(Shulou.com)11/24 Report--

The original title: "there is really ResNet in the brain!" The world's first "Drosophila brain connection Group" was released: it took more than a decade to rebuild 3,000 neurons and more than 500,000 synapses! "

The brain connection group of Drosophila larvae has a long way to go to reconstruct the human brain.

Although modern deep learning has long been separated from the imitation of "biological neural network", understanding the operating mechanism of biological brain is still very helpful for the future development of neural network model.

The structure of brain circuits affects the brain's computing power, but so far, no specific structure of the brain has been seen except in some very simple organisms.

In November last year, researchers from Cambridge University, Johns Hopkins University, Jennifer Research Park and other top institutions uploaded a paper on Biorxiv. After more than a decade of painstaking research, the brain connectivity group of Drosophila larvae was completely reconstructed for the first time.

Link to the paper: https://www.science.org/ doi / 10.1126 / science.add93303 10, the relevant results are published in the journal Science.

One of the authors, Joshua Vogelstein, an associate professor at Johns Hopkins University, said that fruit flies are closer to the human brain in many ways than other creatures, with some areas corresponding to decision-making, some areas corresponding to learning and some areas corresponding to navigation; and the brains of fruit fly larvae and humans are also divided into left and right sides.

In the analysis of Drosophila brain, we can also find some results of modern neural networks, such as cyclic neural network, fast path between multi-layer networks (residual network ResNet), which may inspire the improvement of machine learning model.

The good news: the brain connection group of the reconstructed Drosophila larvae consists of 3016 neurons. The bad news: people have 86 billion neurons.

The complete reconstruction of the brain of Drosophila larvae is mainly composed of "neuronal cells". Adjacent neurons can send signals to each other at the connections between synaptic cells, one of which releases "neurotransmitters" and the other is responsible for receiving this chemical. The complete map of neurons and synapses in the brain is called connectome, which is crucial to understanding how the brain behaves.

The main process of reconstructing the connection group is to cut the brain into ultra-thin (20 micron) slices, and then use the electron flow of an electron microscope to image the slices, such as cutting the brains of Drosophila larvae the size of salt grains into thousands of slices with the slightest error. You have to start all over again.

For some simple creatures, it is relatively easy to build a complete connection group. The first complete drawing is the beautiful hidden rod nematode, with only 302 neurons in the whole body, which was completed in the 1980s.

But so far, researchers have mapped complete synaptic connections for only three organisms, each with only a few hundred brain neurons.

In 2020, researchers at Google and Jennifer Research Park released a 3D model of a Drosophila brain connection group, containing 25000 Drosophila neurons across different cell types and multiple brain regions. But the model is not a complete brain. Even so, the model contains only 1/4 of the 100000 neurons in an adult fruit fly.

Other research groups are studying the connection groups of animals with larger brains, such as insects, fish, mammals, etc., but because of the large number of neurons, the main research method is to partition the brain and study it separately. it makes it impossible to reconstruct areas of the brain that are cross-spatial and interconnected.

The reconstructed complete connection group belongs to the larvae of Drosophila melanogaster, which can show very rich behaviors, including learning, value calculation and behavior selection, and have homologous brain structures with adult Drosophila melanogaster and larger insects.

Powerful genetic tools can be used to selectively manipulate or record individual neuron types, and in manageable (tractable) model systems, assumptions about the functional role of specific neurons and loop motifs revealed by connective groups can be easily tested.

The team cut the brain of a "six-hour-old" Drosophila melanogaster larva into 4841 slices, scanned it with a high-resolution electron microscope, digitized the image and reassembled it into a three-dimensional image; with the help of computer analysis, the resulting map contains 3016 neurons and 548000 synapses.

The residual network is hidden in which researchers analyze the structure of brain circuits (circuit) in detail, including connections and neuronal types, network centers (network hubs) and neural circuit maps.

Most of the brain input-output centers (in-out hubs) are "postsynaptic centers to learning centers" or "presynaptic centers to dopaminergic neurons that drive learning"; map embedding (graph spectral embedding) techniques are used to classify hierarchical clustering neurons based on synaptic connectivity into 93 types, which are internally consistent based on other characteristics such as morphology and function.

The researchers also developed an algorithm to track signal propagation in the brain's multi-synaptic pathways and analyzed feedforward (sensory to output) and feedback pathways (feedback pathways), multi-sensory integration (multisensory integration) and cross-hemispheric responses (cross-hemisphere interactions).

Widespread multi-sensory integration and multiple interrelated pathways at different depths from sensory neurons to output neurons have been found in the brain, forming a distributed processing network.

The brain has a highly circulatory (recurrent) structure, with 41% of neurons receiving long-range cyclic inputs, but the circulation is unevenly distributed, with a particularly high cycle rate in areas involving learning and action choices.

The dopaminergic neurons that drive learning are one of the most common neurons in the brain.

There are many contralateral neurons (contralateral neurons) projecting to the two hemispheres of the brain. They are the entry and exit centers (in-out hubs). They synapse with each other and promote extensive communication between the two hemispheres. The interaction between the brain and nerves is also analyzed.

The researchers found that the target of descending neurons (descending neurons) is a small number of pre-motor elements (premotor elements), which play an important role in motor state (locomotor states) switching.

Conclusion the complete brain connection group of Drosophila larvae will provide the basis for other theoretical and experimental studies of brain function for a long time, and the methods and computational tools produced in this study will promote the analysis of connectors in the future.

Although the details of brain tissue vary in the animal kingdom, many neural circuits are conserved.

As more brain connection groups of other organisms are drawn in the future, comparisons between different connection groups will reveal common and possibly the best circuit structures, as well as the characteristics of differences in behavior between organisms.

Some structural features observed in the brain of Drosophila larvae, including multi-layer shortcuts and significant nested cycles, can be found in the most advanced artificial neural networks, which may make up for the problems of the current network in depth and task generalization. These features can also increase the computing power of the brain and overcome the physiological limitations of the number of neurons.

Future analysis of the similarities and differences between the brain and artificial neural networks may help to understand the computational principles of the brain and may inspire new machine learning architectures.

Gaspar Yekay, a professor of neuroscience at the University of Exeter in the UK, said that all neurons in the brain were reconstructed and all neural connections were analyzed. He called the work "very important".

Katherine Drucker of Harvard University in the United States believes that the data reveal the "deep logic" of these neuronal connections.

But simply mapping synapses doesn't provide the whole picture, says Scott Emmons of the Einstein School of Medicine in New York.

Emmons mapped the connection group of male and female beautiful hidden rod nematodes in 2019, and neurons can also communicate with each other through slow-released chemicals such as hormones and other connections (gap junctions) between cells.

All of this must be taken into account, but only synapses are included in the newly drawn connection group.

Reference:

Https://www.science.org/doi/10.1126/science.add9330

Https://techcrunch.com/2023/03/09/with-this-brain-map-we-are-one-step-closer-to-total-fruit-fly-simulation/

Https://www.npr.org/sections/health-shots/2023/03/09/1161645378/scientists-first-wiring-map-fruit-fly-brain-connectome-human-learning

Https://neurosciencenews.com/neuron-brain-map-22748/

Https://www.engadget.com/scientists-create-the-most-complex-map-yet-of-an-insect-brains-wiring-085600210.html

This article comes from the official account of Wechat: Xin Zhiyuan (ID:AI_era)

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.