As an expert in molecular imaging of single cell signaling in cancer, I develop integrated systems of molecular, cellular, optical, and custom image processing tools to extract rich data sets for biochemical and behavioral functions in living cells over minutes to days. Data sets composed of thousands to millions of cells enable us to develop predictive models of cellular function through a variety of computational approaches, including ODE, ABM, and IRL modeling.
My primary project, election forensics, concerns using statistical analysis to try to determine whether election results are accurate. Election forensics methods use data about voters and votes that are as highly disaggregated as possible. Typically this means polling station (precinct) data, sometimes ballot box data. Data can comprises hundreds of thousands or millions of observations. Geographic information is used, with geographic structure being relevant. Estimation involves complex statistical models. Frontiers include: distinguishing frauds from effects of strategic behavior; estimating frauds probabilities for individual observations (e.g., polling stations); adjoining nonvoting data such as from in-person election observations.
Year
2021-2022
Research Description
Machine Learning Methods for efficient modeling of laser-wakefield particle accelerator (LWFA) simulations and experiments.
Mentor
Alec Thomas, Nuclear Engineering and Radiological Sciences
The schematic is a series of muscle images during the regenerative process, whereby resident stem cells repair the tissue.
Skeletal muscle contains multiple cell types, regenerates when damaged via a pool of resident stem cells (called satellite cells), consumes significant amounts of metabolic energy, grows and adapts its structure and function based on its environment. Nearly all actions in life are defined by the coordinated actions of skeletal muscle but when afflicted by injury or aging, muscle function decreases and quality of life is reduced. Currently, there are little to no therapies for recovery after severe trauma or age-associated muscle wasting (sarcopenia). Prof. Aguilar’s laboratory focuses on studying molecular mechanisms satellite cells use after trauma and aging with a particular focus on transcriptional and epigenetic regulation. We develop and utilize different types of high-throughput sequencing based assays and sophisticated bioinformatics algorithms to generate these insights.
Albert S. Berahas is an Assistant Professor of Industrial & Operations Engineering. Dr. Berahas received a BSC in Operations Research and Industrial Engineering from Cornell University, and his MSC and Ph.D.degrees in Engineering Sciences and Applied Mathematics from Northwestern University. Before coming to the University of Michigan, Dr. Berahas held positions as a Postdoctoral Research Fellow at Lehigh University and at Northwestern University.
Dr. Berahas’ research broadly focuses on designing, developing and analyzing algorithms for solving large scale nonlinear optimization problems. Such problems are ubiquitous, and arise in a plethora of areas such as engineering design, economics, transportation, robotics, machine learning and statistics. Specifically, he is interested in and has explored several sub-fields of nonlinear optimization such as: (i) general nonlinear optimization algorithms, (ii) optimization algorithms for machine learning, (iii) constrained optimization, (iv) stochastic optimization, (v) derivative-free optimization, and (vi) distributed optimization.
Today’s real-world problems are complex and large, often with an overwhelmingly large number of unknown variables which render them doomed to the so-called “curse of dimensionality”. For instance, in machine learning, it is important to obtain simple, interpretable, and parsimonious models of high-dimensional and noisy data. As another example, in large-scale dynamic systems, a long standing question is how to efficiently design distributed control policies for unknown and interconnected systems.
Our research is centered around two main objectives: (1) to model these problems as tractable optimization problems; and (2) to develop structure-aware and scalable computational methods for solving these optimization problems that come equipped with certifiable optimality guarantees. At the core of our research is to show that exploiting hidden structures in these problems—such as graph-induced or spectral sparsity—is a key game-changer in the pursuit of massively scalable and guaranteed computational methods.
Our research interests lies at the intersection of optimization, data analytics, and control.
Josua Stein is an associate professor of Ophthalmology and Visual Sciences at Michigan Medicine. He is the principal investigator of the Sight Outcomes Research Collaborative (SOURCE) consortium, a group of academic ophthalmology departments who are extracting EHR data and data from ocular diagnostic tests for all eye care recipients at their respective institutions, removing all PHI, and sending their data to the University of Michigan where him and his team are cleaning and aggregating the data and making it available to researchers at the various sites for research and Q/I projects. The team is integrating data from SOURCE into machine learning algorithms, applying systems engineering techniques to forecast disease trajectory for chronic eye diseases such as glaucoma, and other precision medicine initiatives.
Dr. Shen is a professor in the department of computer & information science, the University of Michigan-Dearborn, USA. He is a fellow of ASME & IET, and the editor-in-chief of the International Journal of Modelling and Simulation (CiteScore 2018: 1.03), which is an EI-indexed, peer-reviewed research journal published through UK-based Taylor & Francis Group both in print and online. Professor Shen has published over 130 technical papers, 3 books, and organized many international conferences/workshops. His research areas include Modeling and Simulation, Machine Learning and Artificial Intelligence, Numerical Analysis and Optimization, Robotics, Data Science, Sensor Technology, Data Fusion, and Computational Materials Science. Dr. Shen is an innovator who is the creator of two software tools: (a) UM GeoModifier and (b) UM MatDiagnoser. He also contributed to the development of the well-known software OptiStruct.
Intelligent Multiscale Computational Diagnosis of Materials Performance and Life
Kamran Diba is an Associate Professor in the Department of Anesthesiology in the School of Medicine. His research group is interested in how the brain computes, coordinates, stores and transfers information. Neuronal networks generate an assortment of neuronal oscillations that vary depending on the behavior and state of an animal, from active exploration to resting and different stages of sleep and anesthesia. Accordingly, in their recordings of large populations of spiking neurons in rodents, they observe state-dependent temporal relationships at multiple timescales. What role do these unique spike patterns play and what do they tell us about the function and limitations of each brain state? To answer these and related questions, they combine behavioral studies of freely moving, learning and exploring rats, multi-channel recordings of the simultaneous electrical (spiking) activity from hundreds of neurons during behavior and sleep, neural network models of this behavior, statistical and machine learning tools to uncover deep structure within high-dimensional spike trains and chemogenetics and optogenetics to manipulate protein signaling and action potentials in specific neural populations in precise time windows.
Arvind Rao is an Associate Professor in the Department of Computational Medicine and Bioinformatics, and Radiation Oncology in the School of Medicine.
His research is in:
1. Transcriptional Genomics: A bioinformatics framework that identifies tissue‐specific enhancers by integrating multi‐modal genomic data has been developed previously [Rao2010]. There is interest to integrate other sources of information (like epigenomic and ChIP datasets) to improve the efficacy of enhancer prediction. We have also participated in the TCGA Glioma groups’ work [Brat2015, Ceccarelli2016] on identifying transcriptional regulators underlying gliomagenesis.
2. Image Informatics: In order to quantify the phenotypic aspects of disease and their relationships with outcome and their genetic context, we have developed methods for the analysis of histopathology [ Mousavi2015, Vu2016] and radiology [Yang2015] images, focusing on tumor heterogeneity. One direction of our group is to develop image analysis tools to delineate tumor image features from radiology data and to develop predictive models to relate them along with underlying genomic measurements to outcomes in low grade gliomas. Further, we have also investigated methodologies to link tumor imaging, genetics and immune status in gliomas. More recently, my group has been studying the relationship between image-derived features, genetics and cognitive status in glioblastoma patients. Further, we have also developed methods for the analysis of multiparametric MR datasets in Radiation Oncology.
3. Heterogeneous Data Integration: Integrative decision making in the clinical domain involves the need for principled formalisms that can integrate pathology, imaging and genomic data sets to drive hypothesis generation and clinical action. We have focused on developing high throughput measurement pipelines from this diverse array of data sources and methods for their integration. Simultaneously, methods for visualization are also under investigation. A more recent interest of our group is to integrate genomics, imaging and (online) behavioral data from patient to assess their evolving response to treatment, in the context of learning healthcare platforms. This could also enable the development of hybrid diagnostics.
4. Informatics for Combinatorial Drug Screens: the availability of multimodal data sources (cell line genomics, drug assays) coupled with high throughput, high content imaging platforms have created the need for informatics frameworks to identify rational drug combinations capable of modulating disease-associated phenotype. In this context, we have worked with the Gulf Coast Consortium to create analysis platforms that jointly mine imaging and genomics data for combinatorial drug discovery.
The overall goal is to link different data sources, such as imaging-derived phenotypes with genomic alteration for clinical predictive models. This has prompted work in AI/ML models for image processing &computer vision, data integration and genomic analysis.
