How of dynamic functional connectivity (dFC), I saw the

How does the architecture of the nervous system support and constrain the flow of information?
Understanding and reverse engineering a nervous system—the world’s most intricate computational
system—presents unprecedented challenges and opportunities. During my graduate studies in BME at Yale
University, I intend to contribute to this effort by (1) leveraging approaches from network and complex
systems theory to examine multiscale connectomics using multimodal data and (2) developing open-source
tools to improve neuroinformatics infrastructure and reproducibility.
I was first exposed to the intricacies of connectomics research in the neuroimaging laboratory of Dr.
Ajay Satpute at Pomona College. Here, I learned that tools from functional connectomics could identify
ensembles of distributed brain subnetworks—but it was soon clear to me that the picture was incomplete.
Research showed that although experience and behaviour can be highly specific, each subnetwork activated
in response to a highly general repertoire of stimuli. This hinted not only at an ontological problem in the
cognitive sciences, but also at the pluripotency of neural subnetworks.
In the inchoate subdiscipline of dynamic functional connectivity (dFC), I saw the potential for a
finer-grained parcellation with potential to improve our understanding of the structure-to-function mapping.
dFC computes brain connectivity over restricted temporal domains, allowing investigators to track changes
in connectivity over seconds or minutes. My thesis project used k-means clustering and multilayer community
detection to identify metastable patterns of synchronous activity, or brain states. Specifically, I showed that
each of the traditional brain networks could be temporally decomposed into a set of recurrent network
connectivity states, or NC-states. Furthermore, I used my decomposition technique to demonstrate that the
local NC-states within canonical brain networks occurred relatively independently of global states that
spanned the whole brain. I completed my senior exercise with distinction and published my findings in
Scientific Reports.
When I joined the University of Pennsylvania as a neuroimaging analyst, I became keenly aware of a
reproducibility crisis in neuroscience. In 2012, my PI, Dr. Ted Satterthwaite, had characterised a critical
confound in studies of brain connectivity: in-scanner subject movement. Since that initial discovery,
processing strategies that purported to clean movement-related noise had proliferated, complicating the
landscape of image processing. Without consensus in the field as to which strategies were optimal, the
number of options for processing our data was seemingly endless. We took it upon ourselves to
quantitatively benchmark the most common strategies for denoising neuroimages. Our results were
unequivocal and damning: a contentious protocol called global signal regression was consistently effective,
while one of the most frequently used denoising strategies, based on direct estimates of subject movement,
provided little benefit. This work led to a publication in a top neuroimaging methods journal, NeuroImage.
During my work benchmarking denoising protocols, I had written a set of extensible scripts that could
implement any of the 14 evaluated protocols. In the months to follow, these scripts evolved into a large-scale
neuroinformatics software project. Having attended a small undergraduate institution without imaging
facilities, I had developed an appreciation for open science and data sharing, which had empowered me to
complete my thesis. Now, it was my turn to give back—spearheading the development of this open-source
toolbox, I extended its functionality to support multimodal image processing and quality assessment, with the
intention of giving any research lab an easy way to reproduce our benchmarking efforts in their own data. My
research team has used this software toolbox to process the entire PNC dataset, and it has since been
adopted by several large laboratories across the University of Pennsylvania.
As a student at Yale BME, I am interested in contributing to the work of the Magnetic Resonance
Research Center. In particular, I am enthusiastic about Dr. Todd Constable’s research on connectome
fingerprinting, particularly as it relates to changes in brain state. Traditional neuroimaging work has typically
been limited to studying populations and drawing group-level inferences, but studies from Dr. Constable’s
research group and elsewhere suggest that it is possible to make individual-level inferences from
neuroimaging data. With individualised medicine on the horizon, Dr. Constable’s research represents a critical
direction for the future of neuroimaging that accords with my goals. I am additionally interested in
contributing to the work of Drs. Douglas Rothman and Fahmeed Hyder in magnetic resonance spectroscopy
by developing and applying neuroimaging technologies to map neurotransmitter pathways and the laminar
structure of the cortex. More broadly, I am attracted to the multimodal expertise and interdisciplinary
collaborations of the Yale faculty. In summary, I believe that the training opportunities at Yale University are
concordant with my scientific objectives and will prepare me for future leadership in neuroengineering