Neurons, the uber-connected nerve cells that act as a main switchboard for the brain, are central to some incredibly complicated processes. They make it possible to think, walk, speak, and breathe. They even have built-in backup batteries to use in emergencies.
Yet the way individual neurons go about their business is surprisingly simple, according to a new Yale study.
How simple? Most of them operate entirely like tiny on-off switches.
Even the earliest and most basic neuron models going back to the 1940s are still highly accurate, said Christopher Lynn, an assistant professor of physics in Yale's Faculty of Arts and Sciences and author of the new study in the journal Nature Physics.
"I was so surprised. We have 100 billion nerve cells firing in our brain, and each one of them has 10,000 connections to other neurons," Lynn said. "There was no reason to expect that a single neuron has such a simple description."
Lynn, who is also a member of Yale's Quantitative Biology Institute and Wu Tsai Institute, has spent his career studying the ways that neurons combine to form networks in the brain and how complex structures and functions emerge.
For the new study, he switched his perspective, looking instead at the inner workings of individual nerve cells. He developed a computational model that divided neuron activity into three pieces: simple interactions between one input and one output; complex interactions involving multiple inputs; and "latent noise," which arises from the inherent randomness of synapses and reaction spikes.
Neurons communicate via electrical signals, punctuated with tiny releases of chemicals. They are connected by synapses, which transmit electrical signals from one neuron to another.
Lynn applied his model to study the brains of mice and the worm C. elegans. He found that 90% of neuronal activity in the mouse data involved basic, one input-one output interactions. In the worm data, the percentage was 60 to 70%.
"I was expecting to see a roughly equal percentage for each of the three types," Lynn said.
The findings are in line with some of the earliest modeling of artificial neurons dating back to the groundbreaking work of Warren S. McCulloch and Walter Pitts, who produced the first mathematical model of a neuron in 1943. That work proved seminal in the development of artificial neural networks that shaped today's machine learning models.
Lynn said he plans to continue his research on the activity of individual neurons. For example, he wants to compare his new findings with neural data for other animals and learn whether neurons behave differently depending on the complexity of the species.
The work was supported, in part, by the National Institutes of Health.
