Identifying novel associations for iron-related genes in high-grade ovarian cancer
Abigail Descoteaux – Vassar College, Jose W. Velzquez – University of Puerto Rico at Cayey,Anna Konstorum – UConn Health, Center for Quantitative Medicine, Reinhard Laubenbacher – UConn Health, Center for Quantitative Medicine
Iron's ability to gain and lose electrons is enzymatically useful in reactions that promote cell replication, metabolism and growth; however, this property also promotes free radical-generating reactions and thus can contribute to both tumor initiation and tumor growth. It has been found that cancer cells sequester iron by increasing influx into, and decreasing its efflux out of, the cell. High-grade serous ovarian cancer (HGSOC) is the deadliest subtype of ovarian cancer (OVC). In this project we analyzed microarray gene expression data from HGSOC patients in order to build a weighted, undirected network of genes connected by pairwise correlations. The Order Statistics Local Optimization Method (OSLOM) for network community detection was used to identify highly connected subnetworks that contain iron-related genes such as the iron importer transferrin receptor (TFRC) and export regulator hepcidin (HAMP). Biological pathways involving genes in these communities were identified through an over-representation analysis. A surprising interaction was found between iron-related genes and immune pathways. Our algorithm allowed us to generate testable hypotheses regarding the interactions between iron, immune pathways, and ovarian cancer progression.
Network Analysis and Time Series Modeling of Healthcare Data
Lauren Geiser* – University of Dayton, Beverly Setzer – North Carolina State University, Berkeley, Jason Cory Brunson – UConn Health, Center for Quantitative Medicine; Reinhard Laubenbacher – UConn Health, Center for Quantitative Medicine.
Evaluation of patient risk for diseases and adverse health events is important and frequently performed by clinicians. Oftentimes, however, this evaluation is estimated using imprecise and insufficient factors that may not be specific enough to the patient to be as helpful as possible. In the manner of precision medicine, we aim to improve on current clinical risk-evaluation methods by taking into account differences in patient groups not routinely considered in clinical research. We build data-driven mathematical models and modeling pipelines to predict the incidence and severity of diseases based on electronic health records (EHR), concentrating on myocardial infarctions (MI). We use routinely-collected administrative and billing records from the MIMIC-III database, and in a hypothesis-generating sense are interested in what we can learn about heart attack risk from such data. We do so using the generalized linear regression framework and time series models, and explore several subprojects predicting occurrence, in-hospital mortality, and risk progression for MIs. We find that certain diagnoses, demographics, and laboratory tests are useful in predicting these aspects of MIs, despite facing several limitations with our dataset. Predicting the incidence and severity of MIs—and eventually all diseases—can be improved using mathematical modeling with a focus on time series, which can quantify and predict real-world events, and capture changes and relationships between health events over time.
A Diffusion-Based Model of Kinesin Recycling in Neurons
Ryan Fantasia – University of Massachusetts Amherst, Katie Borg – Cornell University, Boris M. Slepchenko – UConn Health, Richard D. Berlin Center for Cell Analysis and Modeling, Masoud Nickaeen – UConn Health, Richard D. Berlin Center for Cell Analysis and Modeling
The axon terminal of neuronal cells has complex signaling and metabolic demands. Transmitting signals to neighboring neurons through synapses involves macromolecules, presynaptic vesicles, and organelles, many of which are synthesized in or near the soma. Since diffusion of large molecular complexes over long distances is slow, and neuronal axons can be meters long, supplying the synapse with cargo from the soma through diffusion would be inefficient. There are motor proteins that actively transport cargo. One is kinesin, which uses energy to walk down microtubule (MT) tracks, pulling cargo along and supplying the needed cargo to the terminus
But what happens to kinesin after transport? Recent experimental work and mathematical modeling have suggested that kinesin may be recycled through a diffusion-based mechanism instead of being degraded or returned via dynein, a retrograde motor as it would according to conventional wisdom [2]. Continuing this line of exploration, we propose a “bucket-brigade” mechanism, in which cargo can “change hands” during active transport, and diffusion of unbound kinesin is restricted by cargo jams, making it no longer a limiting step. These jams arise when active transport is disrupted by spatial inhomogeneities, MT discontinuities or occlusion by microtubule associated proteins (MAPs). We investigate whether this mechanism results in efficient transport over long distances through mathematical modeling using partial differential equations (PDEs).
Dynamic Connectivity in Neural Networks Engaged for Emotional Regulation
Katherine Thai – Rutgers University, Luke Wohlford – University of Arizona, Michael Stevens – Olin Neuropsychiatry Research Center, Institute of Living, Reinhard Laubenbacher – Uconn Health, Center for Quantitative Medicine, Paola Vera-Licona – UConn Health, Center for Quantitative Medicine
Background: Abnormalities in emotional regulation (ER) are implicated in numerous psychiatric disorders, but the connectivity in functional brain networks activated in ER tasks is not well-characterized. Dynamic connectivity in neural networks is a recent neuroscience development that can also be explored in the context of fMRI scans during ER tasks.
Methods: The objective of this project is to construct and validate a pipeline that models fMRI data obtained during ER tasks to probabilistic Boolean networks (PBNs). From the PBNs, dynamic signatures are extracted, then classied into clinically relevant groups by with machine learning. Phenotypic data from patients will be merged with the dynamic signature groups to look for relationships between patient phenotypes and dynamic signature phenotypes. This pipeline will be validated using in silico fMRI data from simTB, a MATLAB toolbox, which has a known prior network structure.
Results: A novel evaluation for data binarization methods was used for both real and in silico data. The mean binarization method was the best for all in silico data sets, but increased variability between real fMRI scans suggests binarizations should be customized for each real patient. The AUC analysis of the static networks reverse engineered from in silico data suggests that two correlation methods are the best performing. Those two methods were combined to form consensus networks for all real and in silico patient data. Preliminary dynamic signatures were calculated for in silico data, but the reverse engineering of PBNs from binarized data and consensus networks proved to be computationally challenging.
Conclusions: The pipeline has been successfully used on in silico data to extract dynamic signatures using PBNs, demonstrating the utility of using such a pipeline. Computational challenges were found with analyzing the real fMRI data, but once this is overcome the pipeline will be benecial in characterizing dynamic connectivity in functional brain networks activated in emotional regulation tasks.
A Model of Iron Metabolism in the Human Body
Timothy Barry – University of Maryland College Park, Mary Gockenbach – University of Texas at Arlington, Jigneshkumar Parmar – UConn Health, Center for Quantitative Medicine, Pedro Mendes – UConn Health, Center for Quantitative Medicine
Iron-related disorders are prevalent throughout the world. Anemia, which has iron deficiency as a major cause, affects nearly one quarter of the world’s population. Hereditary hemochromatosis, a disease of iron overload, is the most common inherited disease of gene mutation in Caucasians. Understanding the mechanisms of iron metabolism in the human body will advance individualized treatments strategies for these and other conditions. A mathematical model using ordinary differential equations is developed to simulate the distribution of iron in the major organs of the body. The model is calibrated for a healthy person using experimental time course data obtained from literature. The inclusion of hormones in the model, such as erythropoietin and hepcidin, enable the investigation of common iron disorders and potential treatments. This model provides a foundation for the creation of a personalizable model in which the specifics of an individual’s condition form the parameter set so that the outcomes of various treatments can be predicted.