Machine learning constructs map of the brain’s neural circuit
Image credit: Afxhome - Dreamstime
Japanese researchers have developed a machine learning model that allows scientists to reconstruct the neuronal circuitry of the brain by measuring signals from the neurons themselves.
According to experts in the field, the brain is considered to be one of the most complex systems in existence. While significant headway has been made to understand how the brain works, researchers tend to generate more questions than answers about this entity.
However, the creators of the machine learning model - a team from Kyoto University - believe it has the potential to explain the difference in neuronal computation in different brain regions more clearly.
To comprehend the brain, neurologists must look at the neurons that construct it. Our entire world of perception runs across these billions of cells in our head and that is compounded by the exponentially larger number of connections – known as synapses – between them. This, therefore, makes the path to our understanding of the brain a challenge.
Shigeru Shinomoto from Kyoto University’s School of Science, who led the project, explained that although it is possible to record the activity of individual neurons in the brain – and that number has increased dramatically over the last decade – it is still a challenge to map out how each of these cells connects to each other.
“It has been suggested that neuronal connectivity can be estimated by analysing the correlation between neuronal signals,” Shinomoto explained. “But getting accurate inference was difficult because of the amount of external noise coming from other neurons.”
As part of their study, the team constructed an analytical method that takes the signal spikes from individual neurons and estimates the inter-neuronal connections from them.
To eliminate data-contaminating 'noise', the researchers applied a generalised linear model (GLM), a basic model in machine learning, to a cross correlogram (an image of correlation statistics) that records the firing correlation between neurons.
“We called our analysis GLMCC and it estimated the strength of nerve connections in units of synaptic membrane potential,” said Ryota Kobayashi from the National Institute of Informatics (NII) based in Tokyo.
“To confirm if our data reflected real-world connectivity, we evaluated its accuracy through a simulation of a large network of neurons. We confirmed that the new model has an accuracy of 97 per cent, much higher than any previous method.”
The model was then applied to experimental data of neuron activity in the hippocampus (a brain structure embedded deep in the temporal lobe of each cerebral cortex) of rats. When analysed, the team found the estimated connections matched the results inferred with other physiological cues.
A 'ready-to-use' version of the deep learning model is available online and the team hopes it will be utilised by neuroscientists around the world.
Shinomoto concluded: “As we advance in technology, the amount of neurological data we collect will increase. Our new analytical model will be vital in processing that information and will lead us to better understand how our brains process the world around us.”
The paper, ‘Reconstructing Neuronal Circuitry from Parallel Spike Trains’, was published in the journal Nature Communications.
In April, researchers from UC Berkeley and the US Institute for Molecular Manufacturing (iMM) predicted that exponential progress in nanotechnology, nanomedicine, artificial intelligence and computation will lead to the development of a ‘human brain/cloud interface’ and will give people instant access to vast knowledge and computing power via thought alone.
In October 2018, E&T explored advances in neuroimaging which could pave way for researchers to observe brain activity in real-time, investigating how these new techniques could reshape the way mental illness is diagnosed and treated.
Sign up to the E&T News e-mail to get great stories like this delivered to your inbox every day.