Skip to main content

Speakers at Northwestern Computational Research Day

The Northwestern Computational Research Day provides opportunities for University faculty, researchers, graduate students, and postdocs to discuss successful practices and challenges in research computing.

Keynote Presentations

Picture of Dr. Desmond Patton

Innovating Gang Violence Prevention with Qualitative Analysis and Natural Language Processing Tool—Louis Room 10:15 a.m. - 11:15 a.m.
Dr. Desmond Patton, Assistant Professor at the Columbia School of Social Work and a Faculty Affiliate of the Social Intervention Group (SIG) and the Data Science Institute

Abstract: Firearm violence continues to be a serious public health problem in the United States. Recent research indicates that victimization and perpetration of firearm violence are exacerbated by social media use and the formation of the “digital street” where youth experiences with violence inform aggressive and threatening content that escalates on social media and leads to retaliation, or Internet banging. Violence prevention and intervention strategies, however, exclude social media as a risk factor, and there are few tools available to community-based organizations for early detection of social media content that may be threatening. Professor Patton will address this critical gap, present his prototype study in which his research team developed a natural language processing (NLP) system, using a small Twitter dataset from a deceased gang member from Chicago. It integrates qualitative analysis and machine learning to automatically identify Internet banging. He will discuss implications for reducing firearm violence using social media.

This is a photo of Dr. Elizabeth McNally

The Omics of Human Genome Sequencing—Louis Room 1:30 p.m. - 2:30 p.m.

Dr. Elizabeth McNally, Professor of Medicine, Cardiology Division and Biochemistry and Molecular Genetics; Elizabeth J. Ward Professorship of Genetic Medicine, Medicine, Cardiology Division, and Director of Center for Genetic Medicine at the Feinberg School of Medicine

Abstract:   Next generation sequencing relies on producing short sequencing reads of 150 base pair in length.  This technology, referred to as massively parallel sequencing, has transformed genetic and genomic analysis by producing millions of base pairs readily, reliably and at comparative low cost.  Genomic profiling is now commonly applied to profile gene expression, genomewide DNA binding protein patterns and histone modifications in a manner where billions of sequence reads are aligned and intepreted.  Since genomic profiling differs for specific cells and tissues, the amount of data from any type of multicellular organism is vast.  It is also now possible to sequence the entire human genome at close to $1000, providing information on both coding and noncoding regions.  These methods have placed new emphasis on computational resources designed to parallelize analysis so that many genomes, or samples, can be analyzed simultaneously and compared.  From this approach, it is now evident that there is more human genetic variation than previously anticipated.  Moreover, with the emergence of large human databases providing genomic sequence data from thousands of racially and ethnically diverse samples, the majority of human genomic information is seen infrequently in the population indicating that humans are each quite unique.  The nature of human genetic variation and its population distribution is changing our understanding of human disease and risk profiling.  Computational resources must be optimally configured to harness the potential of interpreting genomic variation.  

Morning Speaker Sessions

All sessions held at 11:30 a.m. - Noon

Investigating Human Lung Diseases Using Transcriptomics—Northwestern Room A
Paul Reyfman, Fellow, Division of Pulmonary and Critical Care Medicine
Abstract: Chronic lung diseases comprise the third leading cause of mortality in the United States, and the mortality from lung diseases increases with aging. Our group has developed a high throughput approach for performing RNA-seq on key lung cellular populations that have been isolated using fluorescence activated cell sorting. We have used this approach to identify signature of disease in mice during influenza infection and during bleomycin-induced pulmonary fibrosis, and to explore the role of aging in the susceptibility to these diseases. Additionally, we are developing signatures of human disease from transcriptomic profiling of purified cellular populations of lung biopsies obtained at the time of transplantation.  The results of this work can serve as a reference for transcriptomic signatures of disease, and as a basis for evaluating the effects of potential therapies at a transcriptomic level.

Microstructure Evolution in 4D—Northwestern Room B
Yue Sun, Graduate Program in Applied Physics

Abstract: Dendrites are one of the most ubiquitous microstructures formed in various kinds of materials during solidification. They are present in products ranging from engine blocks, gas turbines, to battery electrodes. The morphological evolution of dendritic microstructures during solidification processes has a strong impact on the properties of the final product. The complexity of the dendritic microstructure makes it an interesting yet challenging topic to study. It is even more challenging to study its morphological evolution in four dimensions (three spatial dimensions plus time). We use advanced in-situ experimental technique combined with efficient statistical analysis tools to study the morphological evolution of dendritic microstructures in Al-Cu alloys, and GPU-based computer simulations to validate theoretical models.

Mining Small Data: Gun Violence at Schools Since 1990—Lake Room
Adam Pah, Assistant Professor, Clinical, Management and Organizations, Kellogg School of Management
Abstract: Machine learning, data science, and big data are often how the advantages of computational research is framed. However, computational research can substantially add to the value of small data sets. As an example of this value I will focus on gun violence at US K12 and post-secondary schools. Through analysis of previous datasets on gun violence at schools, we find that there is significant variation in number of identified events of gun violence and temporal trends. Through extensive work, our team constructed a new dataset where individual events were evaluated against a clear set of criteria before being included. Through modelling of the event occurrences, we find that these events have not become deadlier over time; however, there are specific time periods that have different rates of event occurrences. More importantly, these trends in time coincide with the increased emphasis that is placed on post-secondary education over the last 23 years, with more violent events occurring at post-secondary schools from 2007 to 2013.

Afternoon Speaker Sessions

All sessions held at 2:45 p.m. - 3:30 p.m.

Binoculars and bioinformatics: Integrating multiple perspectives to understand the impact of gut microbes on human health—Northwestern Room A
Katie Amato, Assistant Professor, Anthropology, Weinberg College of Arts and Sciences
Abstract: While clinical research aimed at understanding the impacts of the gut microbiota on human health is currently exploding, broader ecological and evolutionary studies of host-gut microbe interactions in wild animals are less common. However, data describing host-gut microbe relationships in selective environments are necessary for putting clinical research into context. My research uses non-human primates both as models for host-gut microbe dynamics in selective environments and as comparative tools for understanding unique aspects of the human gut microbiota and its influences on human physiology and health. In particular, I am interested in the effect of gut microbes on host energy balances. I have used data from wild, black howler monkeys (Alouatta pigra) to develop hypotheses about the influence of gut microbes on host ecology, evolution, and health in the context of nutrition. The human-centered projects I am currently developing directly build on these hypotheses. Here, I will describe the basic patterns I observed in my work with howler monkeys, illustrate how these findings can be incorporated into broader theories of host energetics, and explain how I am currently applying these theories in human research. This work depends heavily on the integration of multiple research tools, including behavioral observations of primates in the field, DNA-based analyses of microbial communities in fecal samples, measurements of host physiology, and an array of open-source bioinformatics tools. Together these approaches promote novel perspectives in a quickly changing field.


Computational Methods for Engineering Design—Northwestern Room B
Wei Chen, Wilson-Cook Professor in Engineering Design, Mechanical Engineering
Abstract: Computational design methods based on mathematical optimization techniques and statistical methods are playing an increasingly important role in designing complex engineering systems and enhancing designers’ productivity in product design and manufacturing. The complexity is considered from various aspects, such as the large number of physical interrelated elements, the complexity and computational cost of design simulations, the heterogeneity of information at different levels of abstraction, the various sources of uncertainties, the multidisciplinary organization with conflicting goals, and the difficulty in understanding the socio-technical interfaces. This talk will introduce the state-of-the-art research in simulation-based design under uncertainty, the use of computational design methods for designing emerging materials systems, and the use of data-analytics in understanding consumers’ preferences to identify important design drivers.


Switching Costs in Pension Plan Choice—Lake Room
Gaston Illanes, College Fellow, Economics, Weinberg College of Arts and Sciences
Abstract: How well do market mechanisms for retirement savings function when there are switching costs? This work answers this question by estimating a dynamic demand model with switching costs for pension fund administrator choice in Chile's privatized pension market. This market exhibits significant price dispersion and very low switching rates, and switching costs are often mentioned as a likely driver of this outcome. If this is the case, then regulatory intervention to lower switching costs may increase welfare. This is not only important for the functioning of the Chilean pension market, but also more generally for other settings where governments mandate consumer participation and set the default as continuing in the same firm as last period. A key challenge in dynamic demand models is the fact that consumers form expectations about the future evolution of product characteristics and base their choices on them. Using a new methodology, based on a combination of revealed preference inequalities and latent variable integration, this work takes these expectations into account without having to model them explicitly, while using exclusion restrictions to separate switching costs from unobserved preference heterogeneity. I find evidence for a lower bound on switching costs of $1,200 dollars, a number significantly higher than that found in previous work. Furthermore, I find evidence that consumers over-value returns differences across pension fund administrators relative to price differences. Observed prices are, on average, roughly twice as high as in a no switching cost counterfactual, suggesting that policy interventions to lower switching costs would be beneficial.

Last Updated: 21 March 2017

Get Help Back to top