Researchers use machine learning to understand synapses
Carnegie Mellon researchers, alongside others, are focusing on neurological research with the goal of gaining a deeper understanding of how specific parts of the brain relate to the organ’s overall function. Currently, many specific functions are thought to be localized in certain parts of the brain. The neocortex, for example, which is comprised of six layers, is the part of the brain that is associated with higher cognitive processes such as spatial reasoning, generation of motor commands, sensory perception, and conscious thought.
Each layer of the neocortex is made up of different cell types, which are organized in a very specific way. Based on a particular person’s experiences, or the “load” experienced by their nervous system, neurons in the brain form or retract, and can even grow new connections. These connections are called synapses, and they form the basis for neural communication. Researchers are currently working on understanding the activity of these neurons and defining the specific set of rules that govern cell-to-cell communication.
A new method developed by researchers at Carnegie Mellon uses machine learning algorithms coupled with a specialized staining technique to create a broader picture of neocortical synapses. This method of microscopy takes in a large piece of tissue and visualizes all of the synapses across its entirety. These synapses appear as little, dark spots with a specific shape.
This method then uses a machine learning algorithm to identify and differentiate the synapses in the data. By doing so, a larger area of the cortical region can be analyzed, compared to the relatively tiny capability of existing microscopy techniques. These techniques not only took a much longer time to analyze the same amount of tissue, but also had a very limited scope of vision. For example, they would allow a “cell-by-cell,” “pair-by-pair connection” to be observed, but the cost, in terms of both time and efficiency, were very heavy. Thus, by giving less importance to cell-type specificity, the team’s new technique allows for a much faster way to understand synaptic activity.
“This is probably the largest microscopy study that has ever been done in terms of looking at lots of animals and a really broad array of tissue,” Alison Barth, one of the key members of the research project and a member of the Center for Neural Basis of Cognition (CNBC) said. The CNBC is a joint program between Carnegie Mellon and the University of Pittsburgh.
Barth explained that this new machine learning technique allows researchers to rapidly count and compare different animals’ brains across different conditions to see what part of the circuit changes when the input is altered.
A study of the data reveals that, when the sensory input that arrived at the brain was changed, the density of synapses across layers changed. The density of observable synapses increased or decreased and in some cases, synapses even became longer. These longer synapses had not been visualized before. Barth explained that “by using machine learning to do the work for us, we are breaking open the direction that we want to go and expanding the areas of the brain that we want to focus on.”
An invention disclosure has been filed for this technique, and Barth believes that there is potential for this technique to be used as a tool for medical diagnosis or understanding learning processes. The study was published in the Journal of Neuroscience and featured on the cover.
The lead author and researcher of this study is Santosh Chandrasekaran, a graduate student from Carnegie Mellon and currently a Postdoctoral Associate at the University of Pittsburgh. In addition to Barth and Chandrasekaran, the research team included Saket Navlakha, currently at the Salk Institute for Biological Studies; Joseph Suhan, a lecturer from Carnegie Mellon’s Department of Biological Sciences; and Ziv Bar-Joseph, a professor in Carnegie Mellon’s Machine Learning department.