Eric Shea-Brown, Ph.D. - US grants

Postdoctoral Research Fellowship

Multi-electrode recording technology and voltage sensitive dyes allow researchers to probe the structure of correlated, stimulus-driven neural activity in groups of cells. However, the diversity of brain areas and stimuli make a complete sampling of these patterns and their effects impossible. Furthermore, the evidence of correlations in stimulus response is strong, yet its role in neural coding difficult to intuit. Therefore, a combined, predictive theory of correlation formation and impact is required. This challenge is approached in three stages. First, a general mathematical theory is developed that relates input correlations of a stochastic forcing to the output correlations of resultant spike trains. The underlying tools will be linear response, population density, and Monte-Carlo methods for the nonlinear stochastic differential equations of spiking neural circuits. Next, this theory is applied to a variety of neural models to quantify how neuron biophysics, morphology, and coupling influence input-output correlation transfer. Finally, information-theoretic analyses are performed to estimate the impact of spike train correlations on the encoding and propagation of sensory inputs in representative neural circuits. Throughout the project, the investigators will work with experimental collaborators to refine and test these predictions.

Understanding the mechanisms by which the nervous system represents and processes information is a fundamental challenge for mathematical biology. It has long been known that information is represented by the intensity of individual neurons' responses. However, new multi-neuron recording and brain imaging techniques are revealing that the information carried by neural tissue is much more (or much less) than the summed contributions of individual neurons. In other terms, the cooperative, correlated features of neural responses can be essential. This poses a pair of fundamental, but unresolved theoretical questions: What are the basic mechanisms by which correlated activity is generated and propagated through layers of neural tissue? What are the consequences for information processing in neuronal networks? The answers will, in stages, make predictions for ongoing neurobiological experiments. For instance, understanding the relation between correlations and neural coding stands to impact the design of neural prosthetics, which code motor and sensory signals via cortical, retinal, thalamic, and cochlear implants. From an alternative perspective, many neurological disorders, such as epilepsy and Parkinson's disease, involve excessive correlation in neural tissue--describing the genesis of correlations and its negative impact on neural coding will aid in designing appropriate treatments that ultimately reduce correlation in the nervous system. Along the way, graduate students involved in this research will receive training in a highly interdisciplinary field, and will gain a broad perspective on mathematical neuroscience through regular visits between three research groups in different regions of the United States. The active involvement of the investigators in undergraduate research and course development will provide an opportunity to translate the questions addressed here into compelling educational topics on the cooperative activity in neural networks that will be accessible to a wide audience.

Questions of neural synchrony -- correlations in cell-to-cell spiking -- have
driven decades of research. However, recent technological and theoretical
advances have thrust open two lines of inquiry. The first is understanding
the combinatorial scale of the correlations that occur in natural and model
neural networks. It is well known that describing neural activity requires
pairwise statistical interactions -- but we do not yet understand when network
dynamics produce patterns of correlations that extend beyond this pairwise
description, when the pairwise descriptions will be complete, and what the
overall implications are for neural coding and signal processing.
The Principal Investigator will address these questions for a set of "canonical"
neural circuits, or motifs, and will build toward networks of gradually
increasing complexity in their dynamics and architecture. Further, extending
beyond intrinsic network dynamics, he will ask how these basic network properties
determine the ways in which patterns of synchrony can be controlled by external stimulation.
Answering these complementary questions requires synergies between methods of
stochastic processes, dynamical systems, and statistical inference.

How do how networked neurons work together to produce the brain's astonishing
computational ability? Such coordinated neural dynamics are characterized by
synchrony among different neurons. One prospect is that this coordinated
activity opens new channels for signal processing: there is a combinatorial
explosion in the number of possible multi-neuron patterns that can occur in
increasingly large networks. However, we only have the first hints at whether
and when these patterns systematically occur in the brain's networks, and what
information they might (or might not) carry. Shea-Brown will study the
fundamental properties of neural dynamics, connectivity, and noise that
determine the level and impact of multi-neuron synchrony in a series of
networks of gradually increasing complexity. He will use
interdisciplinary tools from both deterministic and statistical branches of
applied mathematics to understand how levels of synchrony are created,
destroyed, and manipulated by external stimulation. These findings will
contribute to experimental and clinical neuroscience: working in
collaboration with experimentalists, the investigator will make predictions for light stimuli
that evoke higher-order correlations in the retina, and for electrical stimuli
that suppress pathological synchrony in neurodegenerative disease. These
questions, as part of theoretical neuroscience -- an emerging field that is
rich in open questions and highly varied interdisciplinary techniques --
present a strong opportunity for recruiting, engaging, and training
undergraduates in the mathematical sciences. Shea-Brown will direct this opportunity
toward the underrepresented groups from which new scholars are most urgently
needed, through an integrated four-year research pathway for undergraduates.
This will be developed together with newly designed units in the computational
science and mathematical biology courses taught by the investigator.

New recording methods allow researchers to probe the structure of neural activity with unprecedented scope and detail. As a result there is an explosion of interest in understanding the patterns of activity that emerge in entire neuronal populations and relating these patterns to the function of the nervous system. However, the overwhelming range of different sensory inputs that these populations receive -- and the vast range of different responses that these inputs evoke -- make it impossible to achieve this goal based on empirical observations alone. This challenge is compounded due to the nonlinearity of neuronal network dynamics, which makes it difficult to predict patterns of activity by extrapolation from observations of simpler systems. Predictive mathematical modeling and a deeper understanding of the dynamics of neuronal circuits is therefore required. With previous NSF support, the investigators developed numerical and analytic tools at the interface of statistics, stochastic analysis and nonlinear dynamics, to understand the genesis and impact of correlations in simple, but fundamental microcircuits. They build on these results by extending the underlying mathematical theory to more complex and realistic networks. Using this approach, the team of researchers examines how collective activity is controlled by network architecture, cell dynamics, and stimulus drive in a set of neural networks that typify structures across the nervous system.

Answering these questions will open the door to contemporary biological applications and will meet key theoretical challenges posed by recent technological developments in experimental neuroscience. The key innovation lies in the understanding the collective dynamics of large neural networks that cannot be decomposed into their isolated parts. Through continued interactions with a broad set of experimental collaborators, these ideas are introduced and tested by a broad community of neuroscientists. In the longer term, results on coding in the presence of collective network dynamics will impact the design of neural prosthetics, which code sensory signals via cortical, retinal, and thalamic implants.

Circuits across the brain display correlated activity, in which the responses among neurons are coordinated at a level beyond that expected from the stimulus structure itself. What role does such correlated activity play in neural coding? The retina provides a strong opportunity for progress: widespread correlated activity is generated by key circuit mechanisms that recur throughout the brain. Moreover, the impact of these mechanisms can be studied in the context of stimuli with clear functional significance. Nevertheless, circuit-level nonlinearities at multiple locations cause retinal outputs to depend on stimuli in a manner that defies traditional filter-based models. This demands a tightly coupled computational and experimental approach: purely computational attacks will become lost in the combinatorics of all possible circuit configurations at the difficult "mesoscales" relevant to retinal computation, where averaging approaches fail; purely experimental strategies cannot predict and prioritize the most critical circuit mechanisms and stimulus parameters to explore. The investigators apply this interdisciplinary approach to the mechanisms producing correlated activity in directionally-selective (DS) ganglion cells. In particular, they determine how convergent and divergent pathways in the DS cell circuit interact to shape correlations and encoded information across the cell population, and asses the possible roles of recurrent coupling in modulating this process. These circuit features are the basis of correlated activity in many retinal pathways, and in circuits throughout the brain. Thus, our findings will guide studies of collective neural computation and dynamics currently under intense study in a variety of domains.

The brain translates the sensory environment into its own code, that of neural spikes distributed across vast numbers of neurons. Neuroscience seeks to understand the nature of this code -- what aspects of the spiking activity carry what information, how it is implemented by the cellular hardware of the brain, and how this process can fail in disease. A key puzzle is deciphering what it means when many neurons spike simultaneously -- is this just inevitable statistical coincidence, an artifact of neural hardware, or a key symbol in the code? The investigators take a direct approach to answering this question in the earliest stage of the visual pathway, the retina. Here, the team rigorously combines experiment and theory to connect biological circuit mechanisms and coding, and to identify principles that could be tested in other brain areas. Matching the interdisciplinary demands of this endeavor, investigators from Applied Mathematics and Physiology and Biophysics will unite, and will mentor a small team of undergraduate, graduate, and postdoctoral researchers with diverse backgrounds including both mathematics and biology. They will share their results with peers through an open, user-friendly database, and their most exciting findings with students in the interdisciplinary courses that they teach.

Our brains are constantly changing. Experiences and memories leave their imprints on connections between neurons. Understanding this process is fundamental to understanding how the brain works. While this question has been of central importance to neuroscience for decades, at this moment researchers are well positioned to make significant progress -- new recording devices and imaging techniques are revealing the activity and changes within the networks of the brain at unprecedented scale and resolution. Sound mathematical models are essential to keep up with the mounting avalanche of data. The goal of this project is to develop mathematical tools to assist with improving understanding how networks of neurons are shaped by experiences. Developing this theory is crucial for understanding learning, as well as associated disorders. The project will focus on how learning improves the brain's ability to make decisions and store memories. Graduate students and postdocs joining this project will be part of an established, interdisciplinary mathematics research community. Trainees will gain a wide perspective of mathematical neuroscience through integrated research at three institutions, including extensive visits among them.

This research project builds on earlier results of this team to address a central challenge in the mathematical analysis of biophysically realistic neuronal networks: How brain activity changes brain structure over time. Understanding neural computation demands a description of how network dynamics co-evolves with network architecture. The research team will address this challenge by answering specific questions about the interplay between spatiotemporal patterns of neural activity, the attendant changes in network architectures, and the resulting neural computations. This project focuses on two main questions. First, what mathematical techniques can describe the co-evolution of network dynamics and network connectivity toward stable assemblies of neurons? To address this question this project will build a theory describing how global network structure evolves under the dynamics of biophysically realistic plasticity rules that operate on the scale of individual spikes and synapses. Analysis of these models requires novel multiscale and averaging methods. The resulting equations allow analysis of the stability of network architectures and their dependence on stimulus drive. With these results, the second question can be addressed: How does network plasticity create spatiotemporal dynamics that support the basic building blocks of neural computation? Models to understand how plasticity forms networks whose dynamics underlie specific operations on incoming stimuli will be developed to address this question. The mechanism by which long-term plasticity can reshape the connectivity of a network to encode a precise temporal sequence of events will also be investigated.

Summary This proposal continues and evolves an undergraduate and graduate (NRSA and non-NRSA) Training Program in Neural Computation and Engineering. The University of Washington has a rich history and a large and growing breadth of active teaching and research in this area, with faculty mentors distributed through many departments and schools, including Physiology and Biophysics, Biological Structure, Computer Science and Engineering, Applied Math, Biology, Psychology and Bioengineering. This program evolves the extremely successful previous five-year program which saw the development of an active and highly visible training program, including new undergraduate and graduate program, website, community activities, to take advantage of new opportunities and momentum in Seattle. Support for undergraduate and graduate education and research will enhance interaction between theorists and experimentalists; expand and integrate coursework in emerging approaches in neuroscience, particularly novel offerings in neuroengineering and big data; enhance interactions between undergraduate and graduate students; provide opportunities for undergraduate research and draw together the community across campus to strengthen our already excellent interdisciplinary exchange and collaboration. The undergraduate training program is a 2-year sequence in computational neuroscience, with support for 6 trainees yearly from neurobiology or from a computational/engineering major (Physics, Computer Science and Engineering, Bioengineering, Applied and Computational Mathematics). Trainees take a core curriculum including a research seminar, a choice of laboratory neurobiology sequence and common quantitative courses. Choice of additional electives in an individualized curriculum and career development is guided by a mentoring committee. All students will complete at least 1 and preferably 4 quarters of mentored laboratory research. The graduate training program will support up to 6 students from multiple graduate programs. Students will apply for training grant support at the end of the first year and carry out a core curriculum consisting of neurobiology, quantitative and journal club courses. Individually tailored curricula including electives selected from offerings in computational neuroscience, mathematics, computer science and physics will be devised in consultation with a mentoring committee. Trainees will have access to the UW/Allen Institute Summer Workshop for the Dynamic Brain on San Juan Island. All trainees will attend a regular seminar and and present their research at an annual retreat. The program will be co-directed by Profs. Adrienne Fairhall, Physiology and Biophysics and Eric Shea-Brown, Applied Mathematics, assisted by Leadership Team Prof. Bill Moody, Director, Undergraduate Neurobiology Program and Prof. David Perkel, Director, Neuroscience Graduate Program.

Summary This proposal continues and evolves an undergraduate and graduate (NRSA and non-NRSA) Training Program in Neural Computation and Engineering. The University of Washington has a rich history and a large and growing breadth of active teaching and research in this area, with faculty mentors distributed through many departments and schools, including Physiology and Biophysics, Biological Structure, Computer Science and Engineering, Applied Math, Biology, Psychology and Bioengineering. This program evolves the extremely successful previous five-year program which saw the development of an active and highly visible training program, including new undergraduate and graduate program, website, community activities, to take advantage of new opportunities and momentum in Seattle. Support for undergraduate and graduate education and research will enhance interaction between theorists and experimentalists; expand and integrate coursework in emerging approaches in neuroscience, particularly novel offerings in neuroengineering and big data; enhance interactions between undergraduate and graduate students; provide opportunities for undergraduate research and draw together the community across campus to strengthen our already excellent interdisciplinary exchange and collaboration. The undergraduate training program is a 2-year sequence in computational neuroscience, with support for 6 trainees yearly from neurobiology or from a computational/engineering major (Physics, Computer Science and Engineering, Bioengineering, Applied and Computational Mathematics). Trainees take a core curriculum including a research seminar, a choice of laboratory neurobiology sequence and common quantitative courses. Choice of additional electives in an individualized curriculum and career development is guided by a mentoring committee. All students will complete at least 1 and preferably 4 quarters of mentored laboratory research. The graduate training program will support up to 6 students from multiple graduate programs. Students will apply for training grant support at the end of the first year and carry out a core curriculum consisting of neurobiology, quantitative and journal club courses. Individually tailored curricula including electives selected from offerings in computational neuroscience, mathematics, computer science and physics will be devised in consultation with a mentoring committee. Trainees will have access to the UW/Allen Institute Summer Workshop for the Dynamic Brain on San Juan Island. All trainees will attend a regular seminar and and present their research at an annual retreat. The program will be co-directed by Profs. Adrienne Fairhall, Physiology and Biophysics and Eric Shea-Brown, Applied Mathematics, assisted by Leadership Team Prof. Bill Moody, Director, Undergraduate Neurobiology Program and Prof. David Perkel, Director, Neuroscience Graduate Program.

While the brain's computational abilities are in many respects unrivaled, its workings appear to be far noisier than those of almost any engineered computational system. Even in carefully controlled experiments in which the same conditions are presented over multiple trials, neural activity is strikingly variable from one trial to the next. This project aims to resolve the apparent contradiction between the brain's computational proficiency and its apparently high levels of noise. The core hypothesis is that much of the observed neural variability is driven not by noise but by internal brain modes -- that is, by coordinated patterns of activity across the brain. Thus, variability may be a signature of dynamic and uniquely biological computations rather than noisy fluctuations. If this hypothesis is correct, then observation of these global modes should explain choices that subjects make in behavioral tasks, and perturbation of the modes should alter their patterns of choices in systematic ways. To test this hypothesis, the project employs a novel joint experimental and theoretical approach to measure the variability in brain-wide neural activity across scales, to define its relationship to behavior, and to dynamically perturb these modes to impact behavioral performance both within and across individuals.

To build a transformative understanding of the link between neural and behavioral variability, the project will use multi-probe Neuropixels technology that enables simultaneous recording at submillisecond resolution from thousands of individual neurons distributed across the brain, coupled with advanced data analytic and dynamical modeling tools to extract activity modes from these data. These analyses will be performed together with behavioral assays that probe multiple aspects of behavioral performance, including engagement, perceptual sensitivity, and vigor. Statistical modeling will then be used to identify the functional role of these modes in regulating behavioral performance, and how their activity drives behavioral differences both across trials and across individuals. Beyond correlative analysis, control theory tools will design patterned optogenetic perturbations to provide direct causal tests of this novel functional role for brain-wide activity modes. If the project succeeds, the result will be a new understanding of the nature of the ongoing fluctuations in brain-wide activity patterns trial by trial and individual by individual in terms of behavior rather than noise -- a key step in deciphering the logic of distributed computation underlying perception and cognition. The project will develop and disseminate novel open data and code to impact research and training nationwide. It will also build a dynamic, team-mentored environment led by investigators from very distinct disciplines -- from neuroscience to applied mathematics to control engineering -- which will prepare both undergraduate and graduate trainees to bridge disciplines and scientific cultures.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.

Brains have intricate complexity at many scales, from the detailed structure and connections of neurons to the brain-wide swirl of electrical activity that underlies thought and experience. Making sense of these complexities and using new understanding to design improved computational algorithms requires new combinations of research expertise. This AccelNet project builds international links between academic, private research and industrial partners and prepares the next generation of researchers at the interface between neuroscience and artificial intelligence. The project connects leading centers for academic and industry research in the Pacific Northwest, Paris, France, and Montreal, Canada, to advance understanding of brain structure and dynamics and to use these insights to develop more powerful and efficient computing frameworks that can help mankind to solve the very challenging issues now confronting us.

This catalytic network of networks will spur advances in brain-inspired computation and fundamental theories of neuroscience, particularly with regard to biophysics and connectivity, the role of neuromodulation in creating rich network dynamics, and analytical methods to characterize and control these dynamics. The project combines expertise of researchers gathering brain data from animals engaged in complex tasks, theorists who make models of cognition, researchers who use machine learning to develop new ways of analyzing data, and experts in artificial intelligence. The network of networks will develop a student cohort with the ability to synthesize findings from new and sophisticated data analysis into novel algorithms for computation, and, vice versa, to translate findings from engineering of computational algorithms into hypotheses for brain function. International experience will enhance research opportunities for undergraduate and graduate students. The ethical emphasis of the network will develop a cohort with substantive experience and training in the application of neuroethics and ethical engineering practices.

The Accelerating Research through International Network-to-Network Collaborations (AccelNet) program is designed to accelerate the process of scientific discovery and prepare the next generation of U.S. researchers for multiteam international collaborations. The AccelNet program supports strategic linkages among U.S. research networks and complementary networks abroad that will leverage research and educational resources to tackle grand scientific challenges that require significant coordinated international efforts.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.