Study unravels the wiring of the human neocortex and discovers that unlike mice, human thoughts only flow in one direction
Contrary to what was assumed, the nerve cells in the human neocortex are wired differently than in mice. This is the result of a new study by the Charité – Universitätsmedizin Berlin, which was published in the specialist magazine “Science”. The study found that human neurons communicate in one direction, while in mice signals tend to flow in loops. This increases the efficiency and capacity of the human brain to process information. These discoveries could advance the development of artificial neural networks.
The neocortex, a fundamental structure of human intelligence, is less than five millimeters thick. There, in the outermost layer of the brain, 20 billion neurons process countless sensory perceptions, plan actions and form the basis of our consciousness. How do these neurons process all this complex information? It largely depends on how they are “connected” to each other.
More complex neocortex: different information processing
“Our current understanding of the neuronal architecture in the neocortex is based mainly on findings from animal models such as mice,” explains Prof. Jörg Geiger, Director of the Institute for Neurophysiology at the Charité. In these models, neighboring neurons often communicate with each other as if they were in dialogue. One neuron sends a signal to another and the other neuron sends the signal back. “This means that information often flows in recurring loops.”
The human neocortex is much thicker and more complex than that of a mouse. However, researchers have previously assumed – partly due to a lack of data – that it follows the same basic principles of connectivity. A team of Charité researchers led by Geiger has proven that this is not the case using exceptionally rare tissue samples and state-of-the-art technology.
A smart way to listen to neural communication
For the study, researchers examined brain tissue from 23 people who underwent neurosurgery to treat drug-resistant epilepsy at the Charité. During surgery, it was medically necessary to remove brain tissue to gain access to the underlying diseased structures. The patients had given their consent to use this access tissue for research purposes.
To observe signal flows between neighboring neurons in the outermost layer of the human neocortex, the team developed an improved version of the so-called “multipatch” technique. This allowed the researchers to listen to communication between up to ten neurons simultaneously (see “About the method” for more details). This allowed them to take the required number of measurements to map the network in the short period of time before the cells stopped activity outside the body. In total, they analyzed the communication channels between almost 1,170 neurons with around 7,200 possible connections.
Feed-forward instead of in cycles
They found that only a small proportion of the neurons communicated with each other. “In humans, information tends to flow in only one direction. It rarely returns to the starting point, either directly or through cycles,” explains Dr. Yangfan Peng, lead author of the publication. The team used a computer simulation they developed using the same principles that underlie human network architecture to demonstrate that this forward signal flow has advantages in terms of data processing.
The researchers gave the artificial neural network a typical machine learning task: recognizing correct numbers from audio recordings of spoken digits. The network model that mimicked human structures received more correct answers in this speech recognition task than the model modeled with mice. It was also more efficient, as the same performance required the equivalent of 380 neurons in the mouse model, but only 150 in the human model.
An economic model for AI?
“The directed network architecture we see in humans is more powerful and conserves resources because more independent neurons can perform different tasks at the same time,” explains Peng. “This allows the local network to store more information. “It is not yet clear whether our findings in the outermost layer of the temporal cortex extend to other cortical regions and to what extent they could explain humans’ unique cognitive abilities.”
In the past, AI developers have been inspired by biological models when designing artificial neural networks, but have also optimized their algorithms independently of biological models. “Many artificial neural networks already use some form of this forward connectivity because it produces better results for some tasks,” says Geiger. “It is fascinating to see that the human brain also has similar network principles.” “These findings about cost-effective information processing in the human neocortex could provide inspiration for perfecting AI networks.”
REFERENCE
Directed and acyclic synaptic connectivity in the human layer 2-3 cortical microcircuit.