2017 Winners

Winners of the 2017-2018 Neukom Institute CompX Faculty Grants Program for Dartmouth faculty were announced for one-year projects. We received over $1M in total requests and awarded a total of $225K.

The program seeks to fund both the development of novel computational techniques as well as the application of computational methods to research across the campus and professional schools.

Dartmouth College faculty including the undergraduate, graduate, and professional schools were eligible to apply for these competitive grants. 

Note: An * indicates an award that is partnered with assistance from Dartmouth College Research Computing.

Characterizing the T Cell Receptor Repertoite in Patients with Systemic Sclerosis

Autoimmune diseases are a serious health problem world-wide and in many cases, effective therapies are still lacking. Systemic sclerosis (SSc) is one of the most deadly and difficult to treat systemic autoimmune disease.  SSc is a heterogeneous disease characterized by skin fibrosis, internal organ involvement, and autoantibody production.  SSc presentation among patients is heterogeneous, as some have major internal organ complications that progress rapidly and some are spared these devastating consequences but then live with this chronic disease their entire lives. 

Due to the strong adaptive immune presence in SSc, T cells are likely very important in disease initiation. With recent advances in sequencing technology, it is now possible to leverage T cell receptor (TCR) sequence diversity as a measure of the immune system’s strength and competency within the context of disease progression. We have proposed the first extensive study of TCR sequence diversity and structure in SSc. Our overall hypothesis is that oligoclonal expansion of cross-reacting T cell receptors contributes to the pathogenesis of SSc by initiating autoimmunity via molecular mimicry.

In collaboration with Chris Bailey-Kellogg in Computer Science and Dr. Patricia Pioli in Microbiology and Immunology, we are applying novel genomic sequencing technologies and rigorous computational methods to TCR repertoire analysis to assess genetic changes with changes in protein function related to T cells. We will develop novel computational modeling approaches to identify immunogenic peptides at the TCR-MHCII interface.  This study will be the first extensive, high-resolution study of the TCR repertoire in patients with systemic sclerosis.

One-Handed Text Entry on a Smartwatch using Wrist Gestures

Smartwatches have gained wide adoption and are becoming a major player in the current Post-PC computing era. They are easy-to-carry, always available for use, less obtrusive than smartphones and tablets, and could be fashionable decoration if well designed. However, entering text on smartwatches, especially with one hand only, remains extremely challenging despite the fact that 1) text entry is one of the top activities users perform on small computing devices, and 2) the context of use for smartwatches often requires users to enter text with one hand only: smartwatches are meant to be used on-the-fly where one hand is occupied with a primary task, e.g., walking while holding an umbrella or shopping bags. This problem becomes more significant for blind users as they will have to release guild dogs or put down white canes, which will cause safety concerns in a public environment, e.g. on a bus or in a busy street.

Our goal is to design and develop efficient and easy-to-use one-handed text entry methods on a smartwatch for sighted and blind users. This project will contribute to our long-term vision that anyone can enter text anywhere at anytime, which is crucial to the overall user experience of wearable devices, and is especially important for blind users who often feel left-behind and whose needs of riding technological innovations are overlooked. In the long run, we strive for the future development of very small wearable devices accessible for both sighted and blind users, allowing the input to be unleashed from the “disappearing” touchscreens. Our approach is to use the wrist as an always-available input device to perform one-handed continuous input for text entry. The wrist is one of the most flexible joints in the human body, which can move in mid-air, whirl as a joystick, or tilt to generate continuous gestures. We will develop robust sensors to detect wrist gestures, and create efficient one-handed text entry methods using wrist gestures.

Computational Systems for Geometric Design of Origami Structures

The objective of this research is to develop a next-generation computational framework for the on-demand design of origami structures. Origami, the ancient art of paper folding, features almost infinite possible shapes and has inspired novel designs at all size scales from reconfigurable biomedical devices to foldable space structures, from robotics to furniture. In recent years there has been a burst of interest in origami research due to the richness of the mathematical and physical theories of origami and the promising applications in employing origami structures as building blocks for deployable or rigidly-foldable structures and devices. However, the inverse problem of how to generate crease patterns that are rigidly foldable into a desired three-dimensional shape remains incompletely resolved. Moreover, to design thick-panel origami structures derived from the zero-thickness counterparts and program multistability therein is still a much desired yet challenging task.

The goal is to develop novel computational frameworks to guide the design of tessellation for zero-thickness origami of arbitrary shapes and thick-panel rigid origami structures with multistability and incorporating these frameworks into user-friendly design tools made available to other researchers and the general public. The development of a user-friendly design interface will enable researchers to quickly and cost-effectively determine folding steps and end shapes for a given pattern, as well as to determine the origami pattern needed to produce a given multistable structure.

Developing Computational Tools for Tonal Annotation in Language Documentation

Nearly 50% of the world's languages are tonal, meaning the pitch on which a word is pronounced can contribute to its meaning. Despite its prevalence, tone can be notoriously difficult for researchers to hear and transcribe, particularly when they are not speakers of tone languages themselves. As a result, while linguists been making an unprecedented push to document the world's endangered languages, tonal annotations in these documentary records are often either nonexistent or unreliable, and regardless of reliability, tonal annotation can be extremely time-consuming.

This project aims to develop a computational tool ATLAS (Automated Tone Level Annotation System) to automatically produce time-aligned tonal annotations based on the acoustic signal in the recording. These "phonetic" annotations, reflecting surface-level pronunciation, can be used by future researchers to develop the deeper "phonological" analyses that reflect the speaker's abstract grammatical knowledge.

Associating Fingerprint Patterns with Age and Sex: A Quantifiable Approach

Numerous prehistoric and ancient ceramics, including vessels, figurines, seal impressions, and tablets, preserve the impressions of the fingers and palms of their producers. Scholars working with modern populations have long recognized that certain print characteristics correlate with the sex or the age of the person whose hands produced them.  This project’s goal is to develop a rigorous methodology to evaluate the sex and, when possible, the ages of the producers of ancient Greek clay artifacts, with the objective of eventually evaluating political and social questions like the sex of Minoan administrators or Archaic Greek sculptors. Studies of sexual dimorphism in modern fingerprints can be relatively reliable, reaching rates of accuracy comparable to those available from analysis of skeletal material. However, many of them rely on two-dimensional images of the prints of all ten fingers, and archaeological objects rarely preserve all ten prints.

My hypothesis is that the use of data from the third dimension will allow the sexing and ageing smaller samples of prints. The plan for this year is to use a high-resolution structured light scanner to build a reference sample from the prints left by the genetically and occupationally closest available population, contemporary Greek potters.

The Impact of iPads on Math and Abstraction in Middle School Classrooms

Although many middle schools are implementing 1-1 iPad programs, there has been very little research on the cognitive and educational outcomes of bringing iPads into the classroom. The proposed research aims to examine and better understand (1) how using screen based technologies such as iPads versus traditional pencil and paper impact the way that middle school students learn mathematics, and (2) the ways in which math learners overall use differing levels of cognitive ‘construal’ -- thinking big picture and thinking concretely-- in math learning. In prior construal research, funded by the Neukom Institute, we found that reading the same passage on an iPad versus on paper is associated with lower levels of abstract thinking (2016). Perhaps as a testament to the current concerns about iPads in the classroom, the resulting publication at the 2016 CHI Proceedings received an overwhelming amount of press coverage.

The proposed research continues the examination of the role of technology in abstract thinking and will apply it to middle school math classrooms.

The proposed work will, for the first time, allow researchers and educators to understand how iPads influence mathematical comprehension at multiple grade levels through the technology’s influence on abstract thinking. Funding will allow us to work with schools and teachers to launch a large experiment.

“The Virtual Basilica Project: San Paolo fuori le Mura in Rome”

“The Virtual Basilica Project” is the culmination of fifteen years of work on the Basilica of San Paolo fuori le Mura in Rome. The website to be created will include three-dimensional models, virtual walkthroughs and snapshots that will allow users to visualize nearly two-thousand years of changes to the tomb of the Apostle Paul—one of Christendom’s most venerated sites. The digital models are an accumulation of archival, archeological and visual evidence that has never before been assembled into such rich, composite historical reconstructions.

A Biologically-Plausible Model of (multisensory) Associative Learning

Learning is the process by which experience is transformed into knowledge that can be used to inform future behavior. In the brain, learning is thought to rely on experience-induced changes in synapses, cells, and circuits. Our research strives to determine how these biological changes support the computations necessary for learning. We seek to address these questions by using artificial neural networks to develop a biologically-plausible computational framework for learning, and then conduct anatomical and behavioral experiments to test the central predictions of this framework.

Landscape and Biogeochemical Controls of Methylmercury in Watersheds

In the Northeastern U.S., mercury has built up in watershed soils and sediments from long range atmospheric transport and deposition and point source contamination. Mercury is released to aquatic environments through dissolution of organic matter, and once mobilized, anaerobic microbial communities convert inorganic mercury to methylmercury under appropriate biogeochemical conditions. Methylmercury is a potent neurotoxin that bioaccumulates through aquatic foodwebs, posing a significant exposure risk to humans and wildlife that consume fish. Loading of mercury and organic matter to aquatic ecosystems is predicted to increase under global climate change due to decreased retention in soils from land disturbances and acid rain, and from increased episodes of high runoff. An understanding of the landscapes and processes that drive methylmercury transport in watershed streams, which connect terrestrial sources to larger bodies of water, are needed to predict the fate of mercury under different environmental scenarios.

Landscape, particularly wetland density and forest cover, plays a strong role in predicting the mobility and cycling of mercury and organic matter in watersheds. Once released, the transport and fate of mercury and organic carbon remain strongly tied. Different organic carbon structures have been related to landscape sources, where microbially-mediated components of organic matter are, like methylmercury, associated with wetland inputs, whereas terrestrial, protein-like components are highest during high flow conditions. In this study, we will model the interactive effects of organic carbon compounds and landscape parameters to understand methylmercury loading and bioavailability in streams. We will combine data on methylmercury and organic carbon parameters collected from streams in watersheds across the Northeastern U.S. (Lake Sunapee, NH, Hubbard Brook Experimental Forest, NH; Sleeper’s River, VT; and Arbutus Lake, NY) and obtain landscape parameters (eg. land cover, canopy cover and type) using remote sensing tools. These will be applied to establish a predictive model of methylmercury in streams across watersheds in the Northeast, and provide a framework for investigating the controls on methylmercury levels in streams across different spatial scales.

The Dialogues Digital Project. Landscape Ecology in Central Italy from the Sixth Century to the Present

The Dialogues Digital Project is an interdisciplinary and multimedia interface on Italian landscape ecology which would promote undergraduate research and learning in partnership with professional scholars. Shaped around the Dialogues of Pope Gregory I (c. 600 AD), a series of early medieval miracle stories featuring the Italian landscape, flora, and fauna, this project explores continuities and discontinuities between the socio-political and ecological history of a specific section of Italian territory, a set of multidisciplinary environmental narratives (from c. 600 AD to the present), and local communities. 

Informed by the development of the new field of Environmental Humanities, this collaborative project would provide a digital platform for dialogues between scientists and humanists (both researchers and students), a modeling tool for environmental and cultural awareness, and an opportunity for experiential learning. The two main goals of the project are thus to enrich the specific academic research and experiential learning of students and scholars through interdisciplinary engagement and to produce collaborative research that has the potential for environmental awareness and progress outside the academy. 

The Neukom Institute CompX Grant will support the computational elements of the Dialogues Digital Project by (1) registering all the locations explicitly and implicitly mentioned in Gregory’s Dialogues as well as all references to environmental elements, and to organize the data both geographically and thematically; and (2) constructing an interactive digital map of this territory in which each location mentioned in the Dialogues is characterized by a set of ecological hyperlinks. 

Deep-Learning for Histopathological Characterization of Colorectal Polyps to Improve Colon Cancer Screening

Most colorectal cancer cases start as a small growth on the lining of the colon or rectum, known as a polyp. Although colorectal polyps are precursors to colorectal cancer, it takes several years for these polyps to potentially transform to cancer. If colorectal polyps are detected early, they can be removed before they develop to cancer.

Therefore, screening for colorectal polyps is critical for colorectal cancer prevention. Examination of stained tissue from detected colorectal polyps on glass slides – the practice of histopathology – is a key part of colorectal cancer screening and forms the basis for prognosis and patient management. This histopathological characterization of the detected polyps is an important principle for determining the risk of colorectal cancer and new polyps, and future rates of surveillance for patients. However, examining these slides for colorectal polyp diagnosis is time-intensive, requires years of specialized training, and suffers from significant inter-observer and intra-observer variability. Currently, there is a critical need for computational tools to help pathologists with histopathological characterization and diagnosis of colorectal polyps for more efficient and accurate colorectal cancer screening. This application proposes building an automatic image-understanding method that can accurately detect and classify different types of colorectal polyps in whole-slide histology images. The proposed image-understanding method will be based on deep-learning technology. Deep-learning computational models rely on numerous levels of abstraction for data representation and have shown state-of-the-art results for image classification and object detection, and, in some cases, even exceed human performance.

This method will identify discriminative regions and features on whole-slide histology images that influence the detection and classification results. These regions and their associated features will provide pathologists with insight into the proposed image analysis methodology and will confirm the output detection and classification decisions for the whole-slide images. Upon successful completion of this project, the proposed bioinformatics approach will reduce the cognitive burden on pathologists and improve their accuracy and efficiency in histopathological characterization of colorectal polyps, and subsequent risk assessment and follow-up recommendations. As a result, this project will potentially have a significant, positive impact on improving the efficacy of colorectal cancer screening programs.