One on One with... Tatyana Sharpee
Assistant professor Tatyana Sharpee is among the newest faculty members at the Institute. Since joining Salk in 2007, she has been recognized with several prestigious research awards, including the Alfred P. Sloan Research Fellowship, the McKnight Scholar Award and named a Searle Scholar – honors that are reserved for scientists who have demonstrated innovative research early in their careers with the potential for making significant contributions to biological research. An authority on information theory, Sharpee's team uses a statistical method she developed to decipher how the brain codes and processes information from natural visual stimuli.
I understand you first studied the properties of electrons.
What influenced you to turn your attention to neurons?
The theoretical ideas are similar even though the physical systems are very different. And the closest application between statistical physics and information theory is neuroscience. When I was a little girl, I wanted to be a doctor. But I also greatly admired my grandfather, who was a mathematician/physicist. Working in theoretical and computational biology, in particular neuroscience, provides a way to be connected to the two fields. After completing a thesis in theoretical physics, I joined the Sloan-Swartz program during my postdoctoral studies that provides biology training to physicists.
What fascinates you about the brain?
It's remarkably efficient. For example, my desktop computer uses 60 watts of power and some estimate that the brain uses just 12 watts. Yet the brain is very good at doing analog computations that are very hard mathematical problems such as forming a map of surrounding events that come in through our visual system. So for example, your eyes move three times per second and each time you have a pinhole presentation and yet that's not what we perceive. We perceive a whole continuous field. The algorithms for doing this do not exist, except in the brain and we have yet to find out what they are.
You developed a method for analyzing the brain cells'
response to natural visual stimuli. What type of stimuli
do you use in your experiments and how is it different from
We use a collection of scenes that I took while walking through the woods with a video camera. The reason for using such natural stimuli is that they elicit good responses from high-level neurons whose job is to integrate incoming visual information. When such neurons are presented with simplified stimuli devoid of objects, they respond poorly, sometimes not at all. Our hope is that, although we do not know before the experiments which combination of visual features will drive a particular neuron, by taking scenes from the visual world there will be some features of interest to any visual neuron. A collaborative project with Salk professor John Reynolds' laboratory was recently funded by NEI to both develop new statistical methods and use them to analyze responses of high-level visual neurons.
So if you only take measurements from just one neuron at
a time, how do you know what part of the scene caused it
to fire? How do you know it wasn't a shape, or a color that
trigged the response?
That's why people were shying away from using natural scenes: they are so complex that when you get a spike, you didn't know what actually triggered it. Although you can't make a determination from just one frame, it actually becomes possible when you analyze responses from 10,000 different images, taking into consideration which combination of features produced spikes and which did not. Together with my postdoctoral advisors, we worked to develop an algorithm that can do this correctly regardless of which frames were presented. It is very important with natural stimuli because they are often poorly controlled. To illustrate the main idea of the methods let's say an image contains three elements, and the next image contains these elements in different positions or perhaps even different elements. Then as you do your analysis, you start to see, a-ha, this one caused a spike, and this one didn't when the elements were moved apart. So you start getting statistics of the most likely feature that is associated with a neuron's response. Using advanced statistical techniques, we can create templates for neuron activity from natural visual stimuli.
Does your work focus solely on visual stimuli?
We also use auditory stimuli to learn more about the auditory cortex, which is not as well known or studied as the primary visual cortex. Auditory studies in some ways are more complicated, and some argue that primary auditory cortex is more equivalent to high-level visual neurons than to the primary visual cortex. We know that a given neuron will prefer a certain frequency in the tone or maybe it'll have a sweep up frequency or a sweep down frequency, but that's about it. But through the auditory experiments, we learn about a different brain area and as theorists we try to make parallels between different senses, such as vision or audition, and search for overarching principles.
What does your lab hope to learn by these types
One goal is to learn how signals are integrated in the visual cortex to allow us to perceive objects and shapes. I'm also attracted to theoretical questions of optimization of information transfer in networks. There is an emerging viewpoint that the primary function of many biological networks, either within a cell or between cells, as in neural networks, is the transfer and processing of information. So for example, through phosphorylation a given protein can integrate multiple inputs into a single output that is a graded function of the combination of inputs. At least mathematically, this operation is very much analogous to the integration of signals in the nervous system where a given neuron would sum its synaptic inputs, and produce a spike if the result of integration exceeds a certain threshold. In our recent work (currently in press), we have described how networks of such graded nodes can be set up to convey the maximal amount of information. Assuming the network is optimal, our theoretical analysis also provides a way to infer the strength of interactions with unmeasured parts of the network. This is important because only rarely do we have measurements on a complete circuit in the nervous system. We are looking forward to seeing how these ideas will fare against experimental data.
You grew up in the Soviet Union. Where were you in your
academic life during the dissolution period? And was a
career in science encouraged for women during that time?
That was around 1992, so I was graduating from high school. The official policy of the Soviet Union was equality for all, to quote a slogan from a classic Soviet film "woman is a human being too." So I think the government tried to promote participation of women. But there's the official language and then there's reality. I wasn't always comfortable based on the peer pressure, but I think at least at the early stages of their careers women participation was encouraged.
Was there anyone in particular who encouraged you to get
Yeah, there were lots of people. I had a very good physics teacher in high school, Anatoliy Israilevich Shapiro, who had interesting techniques for teaching students. For example, he would provide very small blank sheets of paper, just several square inches, and give us problems to solve. He thought that having to express ones thoughts on the small sheets promoted creativity when you really had to think about what to write in such a small space. And a problem could be, for example: "Write 10 ways to speed up the drying process of a wet umbrella." He always emphasized that it was better to solve one problem 10 different ways than to solve 10 problems the same way. My family also influenced me. My grandfather and grandmother are mathematicians and my parents are physicists, so even though I thought I had a choice, in reality I didn't. I was gently steered into this field.
You work in a field that's dominated by men. Do you ever
feel the need to encourage other women to consider
a career in computational neurobiology?
I recently gave a presentation for Women Scientists in Action at the Rueben H. Fleet Science Center where I spoke to the girls there, but they are so little. I haven't talked to fifth and sixth graders in like 20 years (laughs). So I was trying to tell them about all of the advantages of being a scientist, but I am afraid that my presentation could have been way out of their interests. But it was a learning experience for me as well, so maybe next time I'll be able to connect better with the girls.
You've received several prestigious fellowships in the last
year. How will you use the funds to expand your research?
At the genetic level, we know the code in DNA is universal whether it's a fly, a mouse or a human, and that's a great success of molecular biology. In neuroscience, we don't even know exactly what the code is, other than the spikes are important. For example, we don't know to what extent the precise timing of individual spikes matters, and the answer might turn out to be different for neurons in different systems or species. So one of the grants is meant to help my lab search for universal symbols in neurotransmission, comparing the structure of neural code in visual or auditory neurons in mammals, birds, and/or flies.
What would finding these universal codes tell us?
Disorders of the nervous system are devastating. Unfortunately, the cures that are available are rather blunt or nonexistent, and often do not take into account the fine scale organization of the nervous system. Understanding details of communication with spikes is a pre-requisite for developing better cures.