NSF funds Houston-based teams’ quest to better understand the brain

Teams will use math, statistics, AI and neuroscience to decipher neural networks

The National Science Foundation (NSF) has awarded a total of $5 million to two Houston-based research teams to develop new tools to better understand the behavior of neural networks.

The range of brain imaging technologies available today presents neuroscientists with a conundrum: They have an unprecedented ability to measure the brain in action — from single cells to entire brain scans and at time scales ranging from milliseconds to days — but they do not have the statistical and mathematical tools to fully understand what they measure.

Researchers will use mathematical models, like this 3-D printed representation of a neural network, to help explain how brain circuits behave. (Artwork courtesy of Xaq Pitkow)

To spur the systematic, interdisciplinary research needed for a comprehensive, transformational understanding of the brain in action, NSF awarded a $4.4 million five-year grant to a team of mathematicians, statisticians, computer scientists and neuroscientists from Rice University, the University of Houston, Baylor College of Medicine and the University of Notre Dame. The grant was part of NSF’s Next Generation Networks for Neuroscience program, or NeuroNex, a component of the BRAIN Initiative.

NSF also awarded a NeuroNex grant of $800,000 to a team led by Rice engineer Jacob Robinson for the study of magnetic techniques to stimulate specific, genetically modified neurons in lab animals without restricting their behaviors.

“The amount of available data has expanded enormously,” said Krešimir Josić, a mathematical biologist at the University of Houston and the principal investigator on the five-year grant. “Now that we have it, the question is, What do we do with it? The important thing isn’t the data. It is understanding what the data mean to us.”

In addition to Josić, who’s also an adjunct professor of biosciences at Rice, the grant research team includes three co-principal investigators with joint appointments at Rice and Baylor: Genevera Allen, associate professor of statistics at Rice and the Neurological Research Institute at Baylor, and Xaq Pitkow and Ankit Patel, assistant professors of neuroscience at Baylor and of electrical and computer engineering at Rice. Additional co-PIs are Notre Dame mathematician Robert Rosenbaum, who earned his Ph.D. in Josić’s lab at UH, and Baylor neuroscientist Andreas Tolias.

Genevera Allen

Allen said the grant will fund 10 researchers — either graduate students or postdoctoral fellows — three in her Rice lab, and her group will interact regularly with groups from other labs. She expects all the students to get together on a monthly basis to share what they’re learning.

Josić agreed that the interdisciplinary nature of the team will be central to meeting the goals of the NeuroNex project. “This is not something you can do in isolation in an office,” he said.

He said the initial work will involve using data gathered from the visual cortex of mice and later move to data captured in more complex situations. Josić and Rosenbaum will study the link between cellular activity and brain function, and they will develop theoretical models to interpret the data. Allen will lead development and validation of the proposed statistical techniques, while Patel, whose work is focused on machine learning, will train artificial neural networks at tasks that parallel those in the experiments. Pitkow will lead the application of graphic models to the analysis of neural activity, stimuli and behavior for artificial and biological neural networks engaged in tasks.

Xaq Pitkow

Allen said the team hopes to create graphical models and other types of models to both explain how neural networks are behaving in recorded tests, as well as to predict how they will behave under specific circumstances. And she said that to help validate whether those models are correct, the team will use maps that Tolias’ group is creating of the physical connections between individual neurons in 1 cubic millimeter of the brain.

“Because every single one of those neurons has connections outside the field of scope to other neurons that you cannot see or image, you’re never going to be able to see every connection,” she said. “Instead, we have to develop different network models that somehow account for everything else that’s going on in the brain. We hope to get creative with developing some new types of models that are specifically suited to this data. And then the next step after that will be to relate what we’re seeing back to stimuli and behavior.”

Ankit Patel

She said the goal is to look beyond individual neurons and instead focus on the interactions within and between neural networks. As an example, she cited research into the behavior of neural networks in the visual cortex of mice.

“We don’t just want to be able to look at the movies and predict what the mouse was seeing,” she said. “We want to know what changed in the network that allowed us to make those predictions. How are the neurons communicating within networks? Because the ultimate goal is really to understand how the brain works.”

The innovation grant to Robinson’s team will develop a technique they call “magnetogenetics.” Their goal is to genetically modify select brain cells so that they respond to magnetic fields that can freely penetrate bone and tissue.

They’ll start small, with flies, said Robinson, assistant professor of electrical and computer engineering and of bioengineering.

Jacob Robinson (left) and Caleb Kemere

“There is much that we still don’t understand about how magnetic fields activate specific cells, but fruit flies provide an excellent test bed for studying magnetic sensitivity and developing tools for neuroscientists,” Robinson said. “Because flies have a rapid life cycle and we have a number of technologies to manipulate the fly genome, we can a create variety of genetically modified flies and study their response to magnetic fields.

“These studies will help us understand how to engineer magnetically sensitive brain cells and lay the foundation for technologies that could remotely control neural activity in model organisms like rats and mice,” he said.

Ultimately, the team hopes that the ability to use magnetic fields to activate or inactivate specific brain cells in freely behaving animals will help reveal fundamental principles of brain function that are conserved across species.

Co-investigators on the award are Caleb Kemere, an assistant professor of electrical and computer engineering at Rice, and Herman Dierick, an assistant professor of molecular and human genetics at Baylor College of Medicine.