2022 Grant Recipients

Winners of the 2022-2023 Neukom Institute CompX Faculty Grants Program for Dartmouth faculty have been announced for one-year projects. The Neukom Institute received over $700K in total requests and awarded $235K of financial support with an additional combination of programming support from Research Computing and the Neukom Scholars program.

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:

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

+ indicates an award that is partnered with an RA provided through the Scholars program

Optical Imaging and Computational Modeling of Electric Field Stimulation of Single Neurons

amanaberra_headshot.jpg

Transcranial brain stimulation techniques allow for the noninvasive delivery of electrical current to the brain, which can be used to activate, inhibit, or modulate brain activity. These techniques allow for causal investigation of human brain function as well as targeted therapies for neurological and psychiatric disorders. However, despite decades of use, their neuromodulatory and therapeutic efficacy can be relatively weak and variable between individuals. A primary obstacle to developing more effective protocols and devices for brain stimulation is the gap in our understanding of the fundamental mechanisms by which single neurons and their microscopically thin axonal and dendritic branches respond to external electric fields. Furthermore, the underlying molecular properties of axons and synapses governing membrane excitability and synaptic transmission are not well characterized due to their relative inaccessibility to conventional electrical recording techniques.

This project seeks to address these challenges by combining optogenetic techniques for resolving the electrical activity and molecular properties of single cultured hippocampal neurons with biophysically realistic computational neuron models. First, we will use genetically encoded voltage indicators to record membrane polarization and excitation by electric fields at high spatiotemporal resolution, allowing us to map the subcellular effects of stimulation across a range of spatial and temporal parameters. Second, we will extend a recently published method for endogenous labeling of neuronal proteins using CRISPR-Cas9 to label and quantify the distribution of voltage-gated sodium and potassium channels in hippocampal axons. These electrical and molecular data will then be used to build and constrain the parameters of compartmental neuron models reproducing the experimental responses to stimulation. This project will provide a mechanistic framework for predicting the direct response of single neurons to electric fields, which is a critical and necessary step in understanding and optimizing the complex effects of brain stimulation on brain networks and behavior.

Augmenting the Organic World: Minimally Invasive Markers

John Bell

Augmented Reality (AR) promises to extend our understanding of our surroundings by inserting digital constructs–sounds, images, text, and 3D models–into real-time camera views of the world around us. Driven by the growing capabilities of smartphones, AR has taken off as a medium for gaming, advertising, and industry training, among other fields. The methods behind these new applications, however, almost always assume an artificial setting such as a house or city where humans have control over its appearance and can use consistent physical structures as reference points to place digital constructs. This CompX project will investigate methods for anchoring AR constructs in an organic setting where visual reference points change rapidly and apply those methods by annotating a forest trail as part of a pilot project: Damiano Benvegnù's Entangled Ecologies.

The project setting requires development of an AR relocalization system that merges data from multiple tracking systems to accurately place digital artifacts. The system is based on "content clusters" containing annotations added to the environment at specific sites. Once the user arrives, the relocalization system will place digital constructs using a combination of GPS, multiple image recognition with custom fiducial markers to establish camera pose, and visual inertial tracking to orient the user when those markers are out of view. By merging multiple tracking methods we can take augmented reality to locations it's never been able to support before.

Drone Lidar 2.0

casana_iraq2019_copy.jpg

Over the past decade, aerial lidar has become a transformative tool for archaeology, enabling the discovery and documentation of vast, previously unknown ancient cities, agricultural landscapes, and ritual installations around the world. Using a low energy laser to scan the surface of the earth, lidar produces high-resolution topographic data, penetrating tree canopy and other dense vegetation and helping to reveal otherwise hidden archaeological features. Yet most lidar data are collected by a costly aircraft-mounted instrument, making aerial lidar surveys extremely expensive to conduct and thereby limiting the ability of researchers to deploy this powerful technology. Thanks to rapid developments in compact, low-cost lidar systems alongside ever more sophisticated drone technology and processing software, it is now possible to collect very high-resolution lidar data using a drone-mounted sensor, but the utility of these new systems for archaeological research remains largely untested.

Building on previous Comp-X supported research, this project will use a new, ultra-compact DJI L1 lidar sensor deployed on the Matrice 300 drone to undertake lidar surveys at several key archaeological sites in forested regions.  We plan to use the lidar system to investigate 1) an ancestral Puebloan settlement at Picuris Pueblo, New Mexico, where stone built houses and agricultural terraces are hidden below pinon-juniper forest, 2) at the Menominee Reservation, Wisconsin, where remains ancient house basins, raised fields, and ceremonial mounds are present throughout forests, and 3) in the Upper Connecticut Valley in New Hampshire and Vermont where both historic and prehistoric archaeological features are obscured by mixed deciduous-evergreen forest.  The project will explore the opportunities and challenges of this new generation of drone lidar technology, revealing many previously undocumented archaeological features and facilitating a range of other projects by Dartmouth researchers.

Discrimination Against those of Asian Descent

Charles Crabtree

To what extent do members of the public and various economic elites (e.g., employers, elected officials, educators, physicians, etc.) discriminate against those of Asian descent in the many economic and social interactions that make up our everyday lives? In the last couple of years there have been many documented instances of hostility against Asians the world over that range from verbal harassment to brutal physical abuse. These have occurred in many advanced democracies including—but not limited to—the United States, France, Italy, Croatia, Finland, Hungary, Ireland, the Netherlands, Russia, Germany, Sweden, Belgium, Canada, and the UK as well as in various developing nations in South America, Oceania, and Africa. With public officials in many nations adding pejorative labels to COVID-19 (e.g. calling it the "Chinese virus") and racial animus on the rise against other racial/ethnic groups the world over, it is vitally important to explore the extent, scope, nature, and origins of hostility against those of Asian descent in economies across the globe.

This project's key objective is to expand our understanding of the scope, nature, and origins of discrimination against those of Asian descent across various OECD countries and, in so doing, to lay the groundwork for expanding the community of academic and non-academic leaders studying and seeking to address this core issue. To achieve this important goal, I will use a unique and novel combination of advanced methods for measuring discrimination in a series of survey experiments, correspondence experiments, and lab-in-the-field experiments to measure discriminatory behaviors against those of various Asian groups—including those from China, Japan, Korea, and other Asian nations. These will provide us with a comprehensive and robust understanding of the nature of bias against those of Asian descent in both high and low-stakes environments. 
 

Textual Moves and the Voices of Others: Longitudinal Research on Student Source Use

Taine Donahue

Tiane Donahue, Isaac Feldman, Nick Van Kley, Sarah Smith, Annika Konrad

Student writers work frequently with written assignments that involve interacting with other texts. The strategies they use and the ways they position themselves in relation to these other voices tell us volumes about their entrance into the scholarly community. Many of the texts students post to their college digital portfolios use and interact with sources, in relation to particular text types assigned and the context of different courses. Our project analyzes the qualitative data of these texts via two quantitative methods: human coding (a social sciences research approach) and NLP-driven automated coding, in order to identify statistically significant trends and put these quantitative results in conversation with our qualitative analyses, notably case study. We will use open source AntConc, and Atlas.ti, for some features analysis, similar to work in big data and digital humanities studies, and grounded in textual corpus analysis which is best adapted to studying change over time, patterns, and comparisons across contexts.

This descriptive study will produce simple analytics (for example, frequency of different text types and textual phenomena); identify statistically significant trends in source use, text use, and use of other kinds of evidence across different text types and course types; track statistically significant change over time in student productions and approaches; and track significant correlations among various contextual factors. The study will inform national conversations about how students work with sources to establish their authority, create strong arguments, and synthesize available knowledge. It will also enable network-building with other US and international teams studying these questions.

Associating Fingerprint Patterns with Age and Sex: A Quantifiable Approach

Julie Hruby

A wide range of prehistoric and ancient Greek ceramic objects, including vessels, ceramic sculpture, seal impressions, and writing tablets preserve the fingerprint impressions of their producers. Traditionally, archaeologists have matched prints in an effort to understand ancient labor systems, but more recently, we have also begun to ask a much wider range of questions. The ages and sexes of producers are among those questions, but so far, the techniques that have been used to reconstruct those factors have typically been able to work on the level of populations rather than individuals, and they have also been subject to challenges posed by differential clay shrinkage rates.

The current project will improve the accuracy of sexing and aging producers of ancient Greek artifacts by using fingerprints that were accidentally impressed in objects made by modern Greek ceramicists as a reference sample. Fingerprint impressions from modern Greek adult potters of known sexes and age grades have already been collected and scanned with a high-resolution 3D scanner, and a Greek attorney has assisted us in complying with both European Union and Greek law as they relate to the collection of prints from juveniles. We have also begun the process of collecting and scanning modern prints from juveniles, and we will begin archiving our raw data.

Hyperglycosylation of ImmunogensviaIn SilicoEngineering (HyperImmunISE)

Jiwon Lee

Glycosylation is one of the most important post-translational modifications of proteins. Enveloped viruses such as HIV-1 and SARS-CoV-2 often exploit the machinery of their host to heavily glycosylate their surface proteins. These glycans can effectively mask various antigenic epitopes on viral proteins to avoid recognition by antibodies, allowing the virus to evade the immune system. We aim to leverage this process to develop an algorithm which introduces non-native glycans on immunogens to block areas on the protein surface and focus immune responses to a particular epitope, with the goal of designing vaccines focusing immune responses to epitopes associated with broad protection. Completion of this proposed study will facilitate in silico protein design and enable the use of these designing tools outside of our research group, as the computational design tool has versatility to engineer many different proteins and will be accessible publicly as a webserver. 

Role of Transfer Learning in the Analysis of Medical Images

Schilling

We seek to answer discrete questions about the role of transfer learning in the analysis of medical images – a necessary next step for advancing the application of deep learning in medical imaging, both in our lab and others.

The advent of deep learning has enabled huge advances in computer vision and thus medical imaging research. The societal benefit of automated interpretation of medical imaging studies, like X-rays, with deep learning is well described.  Broadly speaking, well-trained models have the potential to read imaging studies faster and more accurately while supporting clinicians' workloads and decision processes.  The advent of automated interpretation of medical images also democratizes access to expert interpretation of an imaging study anywhere a digital representation of the study can be obtained.  Since deep learning's emergence, the focus has been on ways to learn faster, with less data, and greater accuracy. The biggest success has been a technique called transfer learning.  Rather than train a computer vision model from scratch, a model developed for one task (e.g. identifying cats in natural images) is repurposed as the starting point for training a model for a different task (e.g., identifying fractures in a radiograph). As a result of the presumed benefits of transfer learning models, it has become the de-facto method for applying deep learning to medical imaging.  Yet, the benefits of transfer learning are not a given when it comes to medical imaging.  That's because successful transfer learning requires that the original and final task share enough similarity. When they don't, final models take longer to train, or lose accuracy.  Natural images (e.g., pictures of cats, dogs, cars, buses, etc.) and medical images (X-rays, MRIs, etc.) differ in fundamental ways and transfer learning's effectiveness is unclear given the imaging domains' unique attributes and tasks. It's possible that transfer learning starting from models trained on natural images offers little benefit to performance, while simple, lightweight models trained on domain-specific images perform comparably or better.  As such, our lab is working to answer discrete questions about the effectiveness of transfer learning in the analysis of medical images – specifically musculoskeletal radiographs.

Gamification of Spoken Data Collection for Sociolinguistic Research

Stanford

Using a novel approach for collecting speech samples for sociolinguistic research, we aim to produce the largest, most geographically comprehensive, audio dataset of North American English dialects ever attempted. The results will have the potential to greatly increase scholarly knowledge of English dialects and vowel systems in general. In this project, we will develop and implement a smartphone app game/research tool designed to be widely downloaded and used for free by iPhone and Android users. We will process the acoustic phonetic data using the online system DARLA, which was created with a prior Neukom CompX grant. Research output will include detailed acoustic analyses of vowel features and statistical correlations with geographic and demographic patterns. The results are likely to provide new, potentially transformative perspectives on vowel systems in North American English.

Rolando

Student Computational Resources for Deep Learning in Digital Pathology

l.vaickus.png

The Pathology Virtual Laboratory is entering its 3^rd year with a new crop of 35 interns. This year we have focused on diversity of applicants at all levels and have recruited from suburban, urban and rural high schools with an emphasis on underserved students. We have 6 students from previous years returning as "near peer" mentors to serve as team leaders. We have additionally added several new faculty mentors.

This research focuses on the integration of quantitative computational techniques into the practice of digital pathology with a focus on deep learning and digital histomorphometry. He is a founding member of the department of pathology and laboratory medicine's new research and educational program: Emerging Diagnostic and Investigational Technologies (EDIT). The EDIT lab's computational research and educational arm is focused on creating a high performance infrastructure for use by faculty and students. DPLM has invested significant capital in acquiring a state of the art computational resource (QDP-Alpha) at the Research Computing Discovery cluster. This resource, combined with the department's vast whole slide imaging and genomics database will enable our researchers and students to perform high quality machine learning investigations in an efficient manner. EDIT is also committed to creating a "virtual laboratory" to allow researchers of all skill levels to fully utilize QDP-Alpha. This resource will provide a secure environment for experimentation and prototyping with robust visualization capabilities. This resource will directly interface with QDP-Alpha to allow for quick deployment of mature algorithms and very large jobs. The Neukom CompX award will allow for the rapid creation of EDIT's virtual lab and ensure that our first crop of remote interns will have world class resources. Dr. Vaickus and the EDIT team are excited to see what our students will create.