Winner of the 2012-2013 Neukom Institute/IQBS CompX Faculty Grants Program for Dartmouth faculty has been announced with awards of up to $20,000 for a one-year project.
Co-sponsored by the Neukom Institute and the Institute for Quantitative Biomedical Sciences(IQBS), the program is focused on funding computational biomedical research in bioengineering, bioinformatics, biostatistics, biophysics or other related areas across the campus and professional schools.
Dartmouth College faculty including the undergraduate, graduate, and professional schools were eligible to apply for this competitive grant. This years winners are:
Currently we are conducting a research on characterizing the spatial distribution of birth defects and low birth weight in the New England, and detect the spatial associations between the birth problems and the arsenic and other environmental exposures from drinking water. So far most birth defects lack medical conclusions on their causes. Certain environmental exposures, especially arsenic, are a recent suspect, but formal epidemiological study is still in its early stage. The research we are conducting is among the pioneering and pilot ones of its kind, especially on an American cohort. An important methodological novelty is that it integrates geospatial analysis with the conventional epidemiological investigation. The exploratory geospatial analysis intends to form hypotheses about the relationships between birth problems and certain environmental factors, and inform further and detailed epidemiological investigations.
This research takes a geocomputational approach based on the Monte Carlo simulation to evaluate statistical significance of the local risk of a birth problem, and to evaluate the spatial associations between a birth problem and an environmental factor. This computational approach is different from the traditional statistical modeling approach that relies on certain statistical assumptions (e.g. the risk of a disease in a population follows the Poisson model). It is also able to take advantage of detailed disease and environmental data that become increasingly available, to quantitatively evaluate the impact of the spatial uncertainties in the data, and to mitigate the impact of subjectivity in the analysis due to data aggregation.
The downside of the approach is that it is computationally intensive. With the current program, it may take unbearably long time to complete an analysis procedure. For example, it took 35 days for a HP Z800 workstation to generate a map of birth defects in New Hampshire. This has seriously limited our capability in experimenting with different data and parameter settings, and considerably delayed the production and lowered the productivity of the research. It has also hindered the adoption of the methods by other researchers in similar studies, although the idea has been well accepted.
With the support of this grant, we plan to revise the program so as to run it on the grid system of the Dartmouth Discovery Cluster. We expect this revision to be able to reduce the running time of an analysis procedure to a fraction and thus considerably improve our research capability and productivity. The outcome of this improvement will certainly not be limited to our current research. Instead, we envision that the revised program will become a core of a larger software package for general spatial epidemiology and public health studies, which may lead to the formation of a research and computation center of the studies of this kind.
Winners of the 2012-2013 Neukom Institute CompX Faculty Grants Program for Dartmouth faculty have been announced with awards of up to $20,000 for one-year projects.
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. This years winners are:
How should a robot efficiently plan a sequence of controls that can be applied to the motors to achieve a given locomotion or manipulation task? This problem of motion planning is central to robotics, and has been studied from many perspectives. The most widely-used and cited algorithms in robot motion planning over the past several years can be broadly classed as sampling-based algorithms. These algorithms sample possible configurations of the robot, and use those samples to build a tree or graph that represents the connectivity of the space of all legal configurations of the robot. The graph is then searched to find a connection between a start and goal.
A central limitation of these sampling-based algorithms is memory. As more computational power becomes available to apply to a motion-planning problem, samples can be generated more quickly. As more samples are placed, the representation of the connectivity of the underlying configuration space tends to become more accurate, but at a tremendous memory cost; if the samples are distributed uniformly over the configuration space, then the number of samples needed for a representation effective for motion planning tends to increase exponentially in the dimensionality of the space. However, the intrinsic difficulty of the problem may not depend on the dimensionality of the space. Planning a trajectory in a very highdimensional space that is obstacle-free may be easy, while a problem in a cluttered lower-dimensional space with tight corridors may prove insurmountable.
We propose to attack the problem of finding low-memory approximate representations of configuration spaces that allow rapid motion planning, and allow questions about connectivity and approximate distances between two points in the configuration space to be answered efficiently. Solving this problem would allow the application of vastly more computational power to motion planning problems, ranging from planning for complex manipulation tasks for robots, to related problems including planning motions for protein folding.
We believe that recent work in the computer science community on data streaming algorithms provides a promising framework for designing limited-memory approaches to motion planning.
The goal of the proposed project is to develop computational models to determine and improve the earthquake response of affordable housing designs for Haiti. Through this project researchers (undergraduate students in a structural analysis course at Dartmouth) will:
Profs. Vicki May and Jack Wilson will work with Thayer students to develop both physical and computational models of affordable housing for Haiti based on the designs developed during the $300 House Design Workshop in January of 2012. These models will be used to suggest improvements to the seismic performance of the designs. Students in Prof. May's Structural Analysis course (ENGS71) in the spring of 2012 will be required to develop computational models and build physical models of housing for Haiti as the course project. The students, primarily 4th and 5th year engineering students, will work in groups to analyze different design options (class size is typically around 35 students). They will be required to predict the frequency response and time-history response of affordable housing options by developing structural models of both the full assemblages of the houses as well as components of the designs (e.g., the walls) using SAP2000. SAP200 allows the user to mix frame elements and solid finite elements to balance model complexity with accuracy and time to run an analysis, an important consideration when running time-history analyses with earthquake input motions. While the 7.0-magnitude earthquake that hit Haiti on January 12, 2010 was not recorded, subsequent aftershocks were recorded and a suite of earthquake motions with a range of frequency characteristics will be used for the analyses. Scale models of the structures built by the students will be tested using a shake table and controller and the response of these physical models to earthquake input motions will be used to calibrate the computational models. Model calibration has been shown to be an important aspect of finite element model development (Sevim et. al., 2011).
|Shown is a heat map of a large microbiome data set collected from infants with cystic fibrosis. On the left shows each sample taken and the bottom shows each microbe identified. The purple color indicates the relative abundance of each microbe, with the darker color indicating larger abundance. These data show that some microbes are conserved in high abundance across samples, while other microbes are restricted to a subset of patient samples|
Over the past several years there has been an increasing recognition in the role of the human microflora (i.e., bacteria and other microbes) in human health and disease. For example, changes in the gut microflora can impact the development of the gut or the likelihood that an animal becomes obese (10). It is also clear that medical interventions (such as antibiotic treatment) have profound and long-term impacts on gut microflora (2, 4). Survey studies have also made it clear that different parts of the body (gut, skin, oral cavity) have distinct microbial populations. It is also not often appreciated that the typical human has 10-fold more bacterial cells than human cells. That is, in terms of numbers, the average human is more microbe than person! There has been an NIH-funded initiative called the Human Microbiome Project (http://commonfund.nih.gov/hmp/) to fund studies in this area of research.
Investigators at Dartmouth have had a long-standing interest in the role of bacterial infections in the lungs of patients with the genetic disease cystic fibrosis (CF). Patients with this disease have one of several thousand possible mutations in a chloride ion channel called CFTR, with the most common mutation a deletion of Phe at position 508 of the protein (1, 3, 7). The CFTR mutation has a number of impacts on the patient, including defects in clearance of bacteria from the lung (5, 6). Thus, CF patients develop chronic polymicrobial infections of the lung that are recalcitrant to even high doses of aerosolized antibiotics. With pilot funds from the Dartmouth Lung Biology Center and the Cystic Fibrosis Foundation, we have initiated studies to assess changes in the microbial populations (e.g., the microbiome) in the CF lung in response to various treatment regimens. In collaboration with Dr. Juliette Madan, we have also begun to explore microbial populations in the gut and upper respiratory tract among developing CF patients from birth onwards, with a current cohort that includes babies that are now 2+ years of age.
A key methodological advance for the study of microbial populations has been the application of deep-sequencing technologies to the study of the human microbiome. Traditionally, microbes have been studied by classical culturing methods, but it has been estimated that in some environments, only 1% of microbes can be cultured (9). In the past two decades, culture-independent based approaches have been utilized wherein a conserved gene, coding for the 16S ribosomal RNA, is PCR amplified using genomic DNA extracted from total microbial samples (clinical or environmental) and the PCR product sequenced. The conserved 16S rDNA gene sequence can be used to generate phylogenetic information that allows the organism to be identified. Using such approaches, organisms present in a sample can be identified without the need for culturing. Initially, such studies involved cloning the PCR products generated and sequencing the clones – typically a maximum of a few hundred clones could be analyzed in such studies using standard DNA sequencing approaches. The advent of deep sequencing technology has revolutionized such experiments by allowing the sequencing of upwards of 50,000 clones per sample. Such a detailed analysis of the microbial populations provides an unprecedented amount of information, including the relative abundance of microbes in the community and how those microbes change in response, for example, to a clinical intervention. Despite this abundance of information, many current microbiome studies simply report which microbes increase and which microbes decrease in response to treatment (very much like the early days of microarray studies). The significance of such changes in microbial populations is often not clear.
Here we propose a new computational approach to extracting information from these microbiome studies. In particular, we propose to exploit the plethora of microbial genome sequence information (several thousand microbial genomes have been sequenced) in conjunction with the availability of the KEGG database (http://www.genome.jp/kegg/). The first goal of this project is to construct a database that scores the presence versus absence of each metabolic enzyme/pathway for each organism we have identified in our microbiome studies. The second goal of this project is to modify a well-developed microarray analysis tool, called Gene Set Enrichment Analysis (GSEA), to determine metabolic properties significantly associated with microbes whose populations change in response to an intervention or as a result of some medical condition. The third goal of this project is to develop and disseminate open-source software based on our methodological developments as a contributed R-package. We believe the computational and statistical tools developed here will mark a significant advance in the ability of investigators to extract biologically significant information regarding why particular microbes respond to interventions, thus providing insight, for example, into how and why therapeutics alter particular subset of microbes but leave other microbes relatively unchanged. Furthermore, using such bioinformatic techniques can allow us to infer properties of the host from which a sample is taken.
Data from a region in Kenya will be used to develop a model of carrying capacity of mosquito larvae as a function of time and weather data. This model will be integrated with epidemiology models of malaria and compared with epidemiology records from the region, in order to establish the validity and potential usefulness of this approach. This project is a prototype for a larger effort for which we will seek funding in the near future.
The goal of this project is to establish a protocol for developing a benchmark data set from a well-studied region in Kenya, against which dynamic models of malaria transmission may be tested. Validation of a model against at least one complete data set is a necessary prerequisite for believing predictions based upon that model. Current progress in malaria modeling is seriously hampered by the lack of a complete data set for any location during any time period. The most difficult aspect of assembling such a data set is determining mosquito dynamics over an extended period, and in particular determining how the carrying capacity for the larval form of the vector changes in time. In this project we will test a prototype for such a study by tracking the physical (hydrological) data that determines larval habitat. We will use this data, along with disease data in five distinct locations over a period of one or more years, establishing a set of measurements that will be a critical resource for those who model not only malaria, but mosquito borne disease in general.
The many dynamical models of malaria transmission already in the literature hold the key to developing optimal combinations of interventions to reduce or eradicate the disease. The major obstacle to this end is the lack of any integrated data collection that would allow these models to be compared, validated, and improved in response to real phenomena. This project will test a prototype solution to this problem through the creation of a somewhat complete data set spanning a one-year period, against which malaria models may be tested.
The Media Ecology project is working to develop digital resources to facilitate Humanities scholarship and teaching regarding historical media. This CompX grant would enable the building of critical tools necessary to realize the goals of the project as a networked, on-line resource. The overall project goal is to develop a (multi)disciplinary applied digital resource derived from Media Studies, to facilitate the awareness of and critical study of Media Ecology: the dynamic ecology of the production, reception, and representational practices of media, in relation to the public sphere and public memory. Our initial content focus is the study of streaming historical news media, particularly newsreels, newsfilm, and television news. Two significant research and teaching outcomes are A) to underscore the significance of such news media to the goal of an informed citizenry, and B) to better understand the role of news media regarding popular memory. Scholars such as Robert McChesney have suggested that the rise of the Internet is one key factor in the collapse of commercial journalism in the U.S. The Media Ecology site would counter such an apparent correlation, heightening meta-criticism of journalism via online capacities for access and collaboration.
We are developing the project in consultation with academic and non-academic organizations, and have commitments from several major archives to establish collaborations via access to significant portions of their repositories. These include WGBH in Boston, the UCLA Film and Television Archive, The Library of Congress, The University of South Carolina, The Peabody Award Archives at The University of Georgia, and The Northeast Historic Film Archive.
Our goal for this grant is to develop a Media Ecology portal that will enable better capacities for enhanced intelligent discovery, collection and annotation of on-line media resources. Media discovery would be facilitated by leveraging current search engines (e.g., dogpile, blinkx, videosurf). The Media Ecology portal would serve as the front end, potentially generating multiple searches from one user request. The portal would then allow the user to capture the location of the media resource (bookmark) and add meta-data (citation, annotation, tags) to the resource location. If allowed and technically possible, existing meta-data from the media resource would be harvested and saved when the resource is located. Subsequent search results could include already discovered resources as part of the new results, and would allow the searcher to provide additional meta-data. Beyond accessing current public resources, we have been in contact with private archives (noted above), and envision extending the reach of the Media Ecology portal into these archives.
Last Updated: 5/15/13