Modeling the brain
October 5, 2015
With more than 86 billion neurons and many trillions of synaptic connections, the human brain is the most complex piece of organized matter in the universe. Few people understand that complexity better than the Modeling, Analysis and Theory (MAT) team at the Allen Institute, who are tasked with making sense of the masses of data collected by lab scientists and ultimately creating theories that can predict aspects of the brain’s behavior.
Computational models are one of the tools employed by the MAT team. These computer simulations are a way to make sense of the physics that underlie the activity of our brains, whether at the level of how a single synapse functions to how entire populations of neurons encode information.
“Models are the crucial link between the data we collect and our theories about how the brain works,” says Nicholas Cain, Scientist I, part of the MAT team. “They provide a scaffold for how to collect the data we’re interested in, and act as proving ground for theories we have about how the brain works.”
Good models have some key characteristics. First, they must match the physical scale of the data. A model of the activity of a couple of synapses looks very different from a model of thousands of neurons.
They also must make reasonable simplifications of the system. “If you look at the organization of the Internet as only a web of connections, ignoring the actual content, you have sacrificed almost all of the substance,” says Cain. “But what you have gained is a manageable model which can be extraordinarily useful despite its reduced form, and can be used to build amazing things like the Google search engine.
“Modeling is ultimately the art and science of striking a balance between usability and accuracy, all for the purpose of hypothesis-driven science,” says Cain.
Writing a program to simulate the brain cannot be done in a vacuum—it needs to be influenced by real-world data in order to be usable. Cain and other members of the MAT team spend much of their time writing software to analyze experimental data, and then finding links between the results of that analysis and existing theories about how the brain works at a more abstract level.
The dream is to be able to articulate the physics of the brain in beautifully simplified language that proves we understand how the entire system works. Cain acknowledges, though, that it is very possible we are too complex to ever reduce the human brain to a simple set of equations.
“At this stage of our understanding, unlike in physics, we can’t write single-line models that capture who we are,” says Cain. “Our current models are many thousands of lines of equations with tens of thousands of variables. But whatever the length, models are important tools for helping us arrange our knowledge as we try to make sense of the brain for the phenomenally complex system that it is.”