One Small Step For A Mouse: Using Information Science to Understand the Brain
by Allison Kleber
How does learning a new skill or process change the physical structure of the brain? Using techniques from data science and high-dimensional statistics, Professors Bobak Nazer (ECE), Venkatesh Saligrama (ECE, SE), and Xue Han (BME), aim to find out. Their project, titled “Discovering Changes in Networks: Fundamental Limits, Efficient Algorithms, and Large-Scale Neuroscience,” has won the support of a $1.2M National Science Foundation (NSF) Award.
Contemporary technology allows researchers to collect enormous amounts of data across a broad range of disciplines, providing them with raw material for the observation and analysis that can lead to new discoveries, technologies, and a deeper understanding of the world. In the course of their recently-funded project, Professors Nazer, Saligrama and Han intend to develop models and algorithms that will allow them to analyze neural datasets collected from the brains of mice, all while working to foster collaboration between the information sciences and large-scale neuroscience.
“Discovering Changes in Networks” is designed to do just that; rather than using a modeling approach that takes raw, “noisy” data and attempts to extrapolate the full structure of a probabilistic graphical model (in this case, modeling the functional connectivity of the imaged neurons), the project focuses on an approach called “network change discovery.” In other words, the proposed algorithms will use the data to determine whether a network’s structure has changed significantly over the course of an experiment, and if so, where. One of the exciting theoretical findings of the proposal is that, if the model structure changes significantly, then detecting this change can be much easier than recovering the network, either fully or approximately. These algorithms will be used to study the networks between neurons in the hippocampi of mice. The team’s goal will be to determine how the (functional) connectivity between neurons in that part of the brain changes over the course of association and instinctive learning experiments; how the structure of those networks is altered as the mice learn a task.
Professors Nazer, Saligrama and Han intend to take three complementary approaches to this problem, drawing upon their respective domains of expertise. First, they will use information theory and high-dimensional statistics to determine the fundamental limits of the process of testing and recovering network changes, in order to understand in what circumstances this change discovery approach is easier than full structural recovery (i.e., when it’s significantly easier to directly detect changes from noisy observations compared to estimating the network structure before and after the experiment, and comparing these estimates).). Next, they will use this framework to design computationally-efficient algorithms, validated first against synthetic datasets, and eventually adapted for the complexities of real data. Finally, these algorithms will be applied to large-scale calcium imaging neural datasets collected from the hippocampi of mice during the learning experiments in question.
The methodology that the researchers hope to produce will have applications beyond their own project; they will also be developing new coursework and open-source resources for fellow investigators in information science and large-scale neuroscience. As part of their proposal to the NSF, Professors Nazer, Saligrama and Han have developed a Broadening Participation in Computing Plan to recruit undergraduate student researchers, with a particular focus on female students. Student researchers—the future of this interdisciplinary field–will have access to cutting-edge techniques and valuable mentorship to guide them on to graduate study, promising careers, and the breakthroughs of the future.