Damian Sendler: Every day, our brains assist us in making myriad decisions, from whether to cross the street to which route to take to the grocery is the most efficient. Our brains are so quick to make decisions, even those that require a lot of thought, that we’re barely conscious of the process involved in making them at all.
The Blavatnik Institute at Harvard Medical School’s assistant professor of neuroscience, Jan Drugowitsch, is intrigued by this mechanism. A neurobiologist with a PhD in machine learning, he studies the brain via a computational lens. Specifically, he is interested in how the brain processes information from the environment and uses this knowledge to influence behavior. With the help of experimentalists, Drugowitsch’s lab tests hypotheses using computer methods.
Doctor Drugowitsch discusses his research on how the brain processes information in order to make split-second decisions in an interview with Harvard Medicine News. He also touches on the necessity of computation and collaboration in solving the mysteries of decision-making.
Damian Jacob Sendler: We’re interested in learning more about your research into the brain and behavior.
Drugowitsch: We spend a lot of time studying how we make judgments based on our perceptions of the world on durations ranging from milliseconds to seconds. The decision to cross the street is an example of an everyday human experience. This requires assessing the current traffic condition and determining whether or not we have enough time to cross the street safely. Most people make this choice unconsciously, taking cues from external factors like traffic flow to the left and right and the sound of approaching vehicles. Our lab studies brain processes like this one, which occur quickly and naturally. We’re trying to figure out how the brain makes these kinds of decisions by combining information from numerous sources throughout time.
Damian Sendler
Our understanding of the process by which we make these decisions has grown increasingly sophisticated over the last few years. This decision-making process follows statistical principles since the information we have is unsure, therefore it’s necessary to compare several sources and ask, “Are we certain enough to commit to a choice?.” This process, with its many intricacies, is being mathematically modelled in my lab using statistical models.
Moving forward, we’re focusing on more ongoing behaviors, like navigation. This is a process that doesn’t have a set of discrete phases; rather, we maintain track of our path constantly and utilize this knowledge to make judgments about our conduct. Our goal is to understand how the brain performs this over time.
Computational methods are an important part of your work. Computational neuroscience is a broad term.
Two types of computational neuroscience are now in use. Computational neuroscience has traditionally used mathematical, physics, and engineering models to describe how the brain performs calculations in assumptions about how it works. These calculations are frequently linked to the way the brain interprets sensory data from the environment. One of the newer forms of computational neuroscience is able to amass significantly more data about the brain. To process the more complicated neurological data, this type of computational neuroscience uses more sophisticated tools that are developed and used. The two are used in our day-to-day work.
Research in my lab focuses on how humans and animals respond to uncertainty in their environments. Our knowledge of the world is essentially unknowable, and dealing with this unknowable information takes us into the domain of statistics. We utilize a lot of statistical techniques because they provide us a good vocabulary to talk about our worldviews and how we see the world. To put it another way, we utilize Bayesian statistics to create models of how uncertainty is handled in the abstract sense. We next utilize physics methods to specify how this statistical information processing can be accomplished in the brain. There are biological limits on how the brain works and how these statistical computations are performed.
HMNews: In your new Neuron study on brain navigation, you make use of some of the aforementioned methods as well. What’s the story behind this project?
An earlier experimental observation about place cells, the cells in the hippocampus of the brain that represent our spatial location, has inspired our research. This is based on what we’ve learned. Place cells suddenly become active in fast bursts while a mouse is standing still, a phenomenon made in mice and rats, and which appears to reflect the animal’s journey through its environment. This activity’s significance is the subject of two competing hypotheses. One benefit is that it aids in the consolidation of previous experiences into long-term memory. In addition, it helps us plan our future routes.
We wanted to better understand the data before tackling these assumptions so that we could better grasp what these bursts do. A more complete picture of activity in place cells was obtained by applying Bayesian statistical methods to previously collected data on rats foraging in a two-meter square area.
Only a tiny percentage of the bursts in place cells were previously assumed to be responsible for stimulating trajectory in open situations. In fact, we discovered that the vast majority of these trajectories contain bursts of activity. Although the animal is motionless, the trajectories of these bursts show velocity as if it were moving through space. However, previous studies have shown that burst activity of place cells during sleep does not contain momentum. Hence, our findings show that, depending on an animal’s state of wakefulness or sleep, bursts of activity in place cells may have a fundamentally different role To better understand how place cells assist us in our daily activities, it is time for us to return to developing computational models.
Damian Jacob Sendler
In your opinion, what are the reasons for the increasing computational nature of neuroscience?
Damien Sendler: Using additional computational tools is in part a response to the wide range of options for obtaining complicated data that are now available. Previously, we could interpret our data without utilizing complicated models if we recorded from a single neuron while an animal performed a simple task. Many more neurons in the brains of animals are now routinely recorded, resulting in data that can only be examined using sophisticated computational models. The awareness that most neuroscientists need at least a rudimentary understanding of how these computer models work has led to a push for more literacy in computational neuroscience.
To that purpose, I co-direct a graduate certificate program at HMS in computational neuroscience for the students. As the need for quantitative abilities among students grew, we realized that the courses we were offering in this area were not sufficiently comprehensive. Developing new courses that teach students how to use all of the computational tools available to interpret neuroscience data is our goal. At the same time, we are trying to improve the cohesion of the computational neuroscience community at Harvard Medical School (HMS).
Computational neuroscience is a fascinating field, but why did you decide to explore it?
As a result of my conviction that comprehending the brain necessitates more complex thinking than intuition alone can provide, I decided to pursue a career in computational neuroscience. I’ve come to believe that we need to construct formal models of how the brain works in order to make progress in our understanding of the brain. By creating these models, we are able to think about brain interactions that are far more intricate than we are able to imagine in our thoughts. We’re relying on the expertise of math and physics to solve this problem.
To a large extent, I’m motivated by a desire to learn new things and uncover the underlying concepts that govern how we do our business. We prefer asking precise questions in our lab since this is the only method to develop predictions that can be put to the test empirically. We do, however, aspire to uncover underlying concepts that underpin these issues. If we are investigating an animal’s behavior, we strive to derive a generalization that we can test in a different set of tests. To answer these problems, computational neuroscience provides us with the necessary tools.
HMNews: You frequently collaborate with neurobiologists from various fields. Why?
For theorists like me, collaborations allow us to work together with talented experimentalists in a mutually beneficial relationship.
This collaboration with experimentalists will allow us to test a number of computational neuroscience theories that have yet to be empirically tested.
Damian Jacob Markiewicz Sendler: Human subjects are sometimes used in tests conducted by scientists. As a result of human experimentation, complicated tasks can be learned quickly. The downside is that it’s difficult to get a clear picture of their minds. We also work with researchers who study animals to answer some of our other issues, such as those pertaining to specific brain connections. Research on the neurophysiology of drosophilas (fruit flies) is one example of our collaboration with Rachel Wilson. Is there a distinct neural circuit in drosophila that performs calculations that we are interested in? These findings may be applicable to other species, including humans, we hope.
In my lab, we may be able to come up with fantastical theories, but at the end of the day, we must be able to connect them to real-world data. We are able to do this because we collaborate with scientists who conduct experiments.
Dr. Damian Jacob Sendler and his media team provided the content for this article.
