Systems Neuroscientists use BRC services to understand the neural basis of perception

Evan Lyall

A childhood filled with Lego blocks, nature documentaries, science fiction novels, and some amazing science teachers led Evan Lyall to know that he wanted to study Bioengineering in college, but left him unsure of the direction he should go with that training. As an undergraduate student at UC Berkeley, Lyall’s interests were honed when he took an introductory neurobiology course, where he was captivated upon “[learning] how neurons convert analog signals to digital signals, how ensembles of neurons are able to wire together to parse out specific features of a sensory environment, and how if you damage specific parts of the brain, you lose specific cognitive functions, such as the ability to form memories, feel fear, or understand language.”  

Now a Biophysics doctoral student studying systems neuroscience under the tutelage of Assistant Professor of Neurobiology Hillel Adesnik, Lyall works with other scientists in the Adesnik Lab to discover “the underlying mechanisms behind perception.” In this pursuit, they focus their investigation on the mechanisms and functions of the cerebral cortex, the outer layer of neural tissue in the mammalian brain; and its fundamental computational unit, the cortical microcircuit, a “wiring of cortical neurons” between and across layers of the cortex.

With other members of the Adesnik group, Lyall has been utilizing Berkeley Research Computing (BRC) computational and consulting resources to advance the group’s research. Savio, the campus shared High Performance Computing (HPC) cluster, is an essential tool in analyses of data-dense videos of neural activity collected by Lyall and his colleagues from the brains of awake, actively sensing transgenic mice, and “has allowed [them] to speed up [their] analyses by an order of magnitude, from multiple hours to a matter of minutes,” according to Lyall. With the help of BRC consultants, Lyall and the Adesnik group are now striving to “analyze this data in near real-time” by transferring experimental data to Savio as soon as it is collected, and using analytical results from Savio’s computations to perform subsequent experimental steps -- an important development for future research. 

The Cerebral Cortex and Cortical Microcircuit 

An “evolutionary development found solely in the mammalian brain,” the cerebral cortex in humans is a highly complex, folded mass of neural tissue, and plays an essential “role in nearly every aspect, if not all aspects, of our conscious brain,” including “working memory, sensory processing, voluntary motor control,” and perhaps “consciousness itself,” as Lyall says. But how is this role performed? The cortex converts data input to information output via the organized and “sophisticated interplay of millions of cortical neurons.” The cortex takes in sensory information from the world outside of us -- for example, the touch of a hand on our shoulder, or the visual stimulus of bright sunlight -- and then “[performs] various computations on that information.” Information representing the world is relayed through the brain as electrical signals that propagate from neuron to neuron, creating a cascade of neural communication. These signals are transformed along this cascade and then are sent to other brain areas for further processing, and thus the “percept” -- a neurally encoded model of an external stimulus that the conscious mind understands -- is developed and ultimately interpreted.

The cortex, as Lyall emphasizes, is a compartmentalized structure, in which different regions accomplish different tasks. For example, it’s well known in the field of systems neuroscience that the primary visual cortex is at the back of the brain, and is essential for visual perception. In addition to this regional functionality, the cortex is layered: the cortex has between five and six layers of distinct neural tissue. Between and within these layers is a communicative array, or “wiring backbone,” of cortical neurons called the cortical microcircuit, a neural network motif that is essential to our ability to perceive and comprehend -- and “evolutionarily fundamental to our intelligence,” as Lyall says. 

How the cortical microcircuit “functionally pulls out features, performs associations, and in general performs the computations necessary to generate downstream perception,” is poorly understood. In the effort to better understand this process, Lyall and the Adesnik group are working to deepen their “knowledge of the functional wiring of the cortex, and how the representation of information changes across each step of the hierarchy,” that is, as the signal encoding the stimulus passes through the cortical microcircuit. 

The lab’s research focuses on the primary somatosensory cortex and visual cortex, which interpret touch and visual information respectively. Lyall’s work leverages the masterful spatial acuity that mice derive from their whiskers to examine the functions of these specialized areas of the brain. He studies “how touch information propagated from a mouse’s whiskers during active sensation is transformed across the layers of the primary somatosensory cortex,” a region known as S1. Lyall explains that to generate a model of the environment around them, mice, as subterranean and nocturnal animals, rely on moving their whiskers back and forth, brushing across, and thereby probing, their surroundings. This is analogous to how a human individual with a visual impairment may run their hands over a surface to examine and understand its features or over Braille to sense the characters of the language. The S1 region is well studied, according to Lyall, but predominantly in anesthetized mice. “This is a very different brain state from that of an actively sensing mouse,” Lyall says. “If we want to understand how the cortex performs the computations that it does, we first have to characterize the representation of the information at the different levels of the system in an ethologically relevant brain state,” meaning when the mice are awake, and exhibiting behaviors similar to those exhibited in a natural environment. 

To capture this data in actively sensing mice, Lyall’s mice are surgically fitted with tiny windows over the cortical region being imaged, through which videos are acquired with a two-photon scanning laser microscope as the active neurons in the specified brain region fluoresce. The mice are transgenic, modified and bred to express a genetically-encoded calcium indicator in their neurons that “emits fluorescence when...there is a significant concentration of calcium within the cell.” A concentration peak occurs when the neuron “spikes” (is activated) and therefore a large electric signal is passing through it. Neural activation, imaged primarily in the upper layers of S1, is induced as the experimenter presents a thin, vertical bar to the mouse whiskers. Delicately, the mouse’s whiskers flutter around the bar. And in the once dark sensory processing regions being imaged, there is light. 

Parallel Computation and Real-Time Data Analysis 

The neural activity videos Lyall collects aid his and his colleagues’ efforts to, for example, “identify recurring patterns of activity encoding different sensory features,” and the Adesnik Lab’s larger effort to “derive a set of generalizable principles underlying cortical computations.” In identifying these patterns and principles, “the services offered by BRC have become extremely essential,” Lyall says. “The videos of neural activity that we acquire are very large and as a result very computationally intensive to analyze,” however with the Savio high performance computing cluster, this data processing can be accelerated through two modes of parallelization. First, Lyall can “parallelize [his] analysis over more CPU cores than are available on local machines.” While his desktop only has 6 cores, Lyall says “a single node on Savio has 20 cores at a minimum,” and further, “a single dataset can be processed over multiple nodes working in concert.” Second, “Savio allows for running multiple jobs at once,” giving Lyall the option to “analyze multiple datasets at the same time,” and, importantly, freeing up his desktop for other tasks. 

“Additionally,” Lyall says, “we have recently begun working with Maurice Manning, a consultant at BRC, who is helping us to design a pipeline to analyze our [videos of neural activity] in near real-time.” This accelerated information flow compliments a new method of experimentation for the Adesnik group: the lab has recently advanced techniques that allow them “to activate or inhibit neurons with cellular resolution and millisecond precision” using temporally focused digital holography. While optogenetics has already given neuroscientists the power to manipulate neural activity using light, and thereby uncover important information about neural coding, the distribution of manipulation is often broad, “generally affecting hundreds to thousands of neurons simultaneously,” Lyall says. The lab’s new technique overcomes this limitation, and with the advent of real-time analysis, will yield a “brain-machine interface that can both read-out and write-in neural activity at single cell and single spike resolution.” This precise control may allow Lyall and his colleagues to even “sculpt the neural basis of the percept as it develops,” reorganizing which neurons are activated in the specified brain region, and when they spike, while assessing “if the mouse can still perceive the stimulus by reading out its behavioral response.” According to Lyall, the real-time data analysis pipeline developed through BRC consultation “will open many doors for us by allowing us to perform pivotal experiments and manipulations that rely on knowing the brain’s state at any given moment.”

On the importance of research computing resources for his work, Lyall remarks, “the realm of systems neuroscience is collectively moving to the acquisition and analysis of larger and larger datasets. The more neurons we sample from at any one time, and the faster we sample them, the more knowledge we can potentially glean about how the neurons are working together to encode and process information at the millisecond time scale. With this in mind, high performance computing has become essential for our field. In this sense, we will be fully reliant on the services provided by the BRC in our pursuit of understanding how the cortex functions and performs the amazing feats that it does.”

If you would like to learn more about using Savio, or consulting with BRC about your research, please e-mail