Researchers from the Allen Institute in Seattle, together with collaborators in Japan, have created a highly detailed supercomputer simulation of the mouse cortex. They modeled nearly 10 million neurons with 26 billion synapses on Japan’s Fugaku supercomputer. Their simulation captures not just the broad structure, but also sub-cellular details: each neuron is represented as a tree of multiple interacting compartments. The program, called Neulite, was able to simulate one second of real-time brain activity in about 32 seconds of computing time, only about 32x slower than a living mouse, which is remarkable for a model of this scale and complexity.
Although this achievement is a major technical milestone, the scientists emphasize that it’s still a long way from modeling a full and biologically realistic brain. Their current simulation lacks important features like plasticity (how neurons rewire themselves) and neuromodulators (molecules that change how neurons behave). It also doesn’t yet capture detailed sensory inputs. The long-term ambition, however, is to simulate an entire brain and not just the cortex. For reference, while the simulated cortex has about 10 million neurons, a full mouse brain would have around 70 million, and a human cortex alone contains around 21 billion neurons.
More information: