Modeling the brain’s connections

May 26, 2016

Scientific discoveries can emerge from unexpected places. A picnic table nestled in the woods of the San Juan Islands, illuminated by the late summer sun rather than the glow of a lab bench lamp, may not seem the most likely place to uncover fundamental properties that govern how the brain is wired. But at the Summer Workshop on the Dynamic Brain in the summer of 2014, this was exactly the case.

undefined“The three of us had never worked together before the workshop, since we’re in different graduate laboratories with one of us studying on a different continent,” explains Mark Wronkiewicz, a graduate student at the University of Washington, who worked collaboratively with Sid Henriksen of Newcastle University and Rich Pang also at the University of Washington on a project that opened all of their eyes to the wealth of data available in the Allen Institute’s publicly available resources. “We all went around the room to figure out who was most interested in which topics, and all three of us were interested in graph theory but none had used it before.”

Graph theory uses nodes and connections to map out the structure of networks, like the ones that connect different regions of the brain. Graph theory has been used to describe smaller networks in the brain in the past, but before these three students conspired on a patio at Friday Harbor, it had rarely been applied to a dataset the size of the entire Allen Mouse Brain Connectivity Atlas.

First, the team dug into the data to discover some fundamental properties of the brain’s connections. One of the first characteristics they noticed was that the direction of information flow in the brain was important, meaning that the connection from point A to point B was different from the connection from point B to point A. This key distinction is often glossed over in more simple “undirected” models of the brain.

The number of connections a particular brain region had to other regions was another key characteristic. If a brain region was only connected to a few different regions, those connected regions also tended to be linked to each other, making for tighter networks.

Using these basic principles gathered from the data, Henriksen, Pang and Wronkiewicz created models that they hoped would replicate these (and other) fundamental aspects of the data in the Allen Mouse Brain Connectivity Atlas.

They programmed their model with two simple rules. The first was that brain regions with many outgoing connections were more likely to form new outgoing connections. “It’s like adding a flight between two cities,” explains Wronkiewicz. “If a city is already a hub with many outgoing flights, it’s more likely to be the start point for a new outgoing flight.” Second, those new outgoing connections were more likely to connect to nearby regions than distant ones.

Although simple, the resulting model guided by these two principles successfully captured many of the characteristics of brain connections represented in the original Allen Mouse Brain Connectivity Atlas data. Their findings were recently published in eLife, an open access journal.

“It was definitely surprising that we stumbled upon something that fit the data so well—a trio of grad students working together for the first time,” says Wronkiewicz. “It comes back to the quality of the dataset, which was a huge contributing factor, and the incredible opportunity the workshop gave us to learn together. We all pushed each other above and beyond what we would usually do in school because the environment was so stimulating.”