Browsing by Author "Nishimura, Erin O., committee member"

Now showing 1 - 2 of 2

Open Access
AI for personalized medicine
(Colorado State University. Libraries, 2023) Lewis, Aidan Michael, author; Bailey, Susan M., advisor; LaRocca, Tom J., committee member; Nishimura, Erin O., committee member
In 2021, Americans spent an estimated $4.3 trillion dollars on healthcare, an extraordinary amount for treatment that is often less effective than care in other developed nations (1-3). Precision, or personalized, medicine represents a new frontier in healthcare that promises treatment plans and optimized health strategies tailored to an individual (4) thereby making medicine more effective and less costly. Contemporary Artificial Intelligence (AI) and Machine Learning (ML) approaches have tremendous potential to advance the field of precision medicine by leveraging the technology's power of deciphering patterns in the data to make predictions about an individual's health outcomes (3, 5-8). However, many developing AI/ML approaches to precision medicine have not proven particularly successful in making accurate predictions and conclusions mostly due to the limited availability of high-quality medical data for input. The Wake Forrest University Non-Human Primate Radiation Late Effects Cohort (NHP RLEC) provides an unprecedented opportunity to test AI's ability to be trained on a comprehensive human health analog data set in an experimentally irradiated Rhesus monkey cohort with extraordinary life-time records of biomarker levels and health outcomes. Here, we test prevalent, scalable ML models to improve accuracy of predictions specifically related to radiation biomarkers, dose, and health outcomes. We find that CatBoost, ElasticNet, and XGBoost models can accurately predict lymphocyte counts for both monkey populations and individual monkeys. Furthermore, these models can accurately predict radiation dose and biomarker levels using only five other features within the models. Although the models were only marginally successful at predicting lymphocyte counts using radiation-related data alone, and at predicting the health outcomes of the monkeys, these findings and this unique dataset represent key steps toward refining the combinations of factors necessary for the successful use of AI/ML models in precision medicine for humans.
Open Access
Live-cell imaging uncovers the relationship between histone acetylation, RNA polymerase II phosphorylation, transcription, and chromatin dynamics
(Colorado State University. Libraries, 2023) Saxton, Matthew Neeley, author; Stasevich, Timothy J., advisor; Nishimura, Erin O., committee member; Hansen, Jeffrey, committee member; Krapf, Diego, committee member
Living cells are capable of turning a one dimensional strand of nucleic acids into a functional polypeptide. A host of steps and factors are involved in the process of transcription and translation, and understanding each of them is necessary for comprehending and characterizing life. While new technologies and assays have expanded our understanding of eukaryotic transcription, there is still much to be learned. In particular, single-molecule microscopy provides a powerful and versatile platform for studying the genesis of RNA with unparalleled spatiotemporal resolution (Chapter 1). First, we characterize the timing, kinetics, and occupancy of phosphorylated RNA polymerase II (RNAP2) using a single-copy HIV-1 reporter system. This work provides strong evidence for clusters of phosphorylated, initiating RNAP2 which is spatially separated from bursty, downstream RNA synthesis. It is found that RNAP2-Ser5-phosphorylation (Ser5ph) precedes RNA output by ~1 minute, and RNAP2 arrives at the locus in a phosphorylated state (Chapter 2). Then, we examine the spatial correlation between H3K27 acetylation and Ser5ph in living cells on the course of minutes to hours. Contrary to expectations based upon ChIP data, we find that the two signals are in fact spatially separated. This argues for a functional separation between transcriptional poising and initiation, likely aiding bursty behavior. Next, the dynamics of single chromatin-incorporated nucleosomes in the context of H3K27 acetylation and transcription initiation is determined with super-resolution single-molecule imaging. The physical movement of chromatin inside of H3K27ac and RNAP2-Ser5ph enriched regions is found to be significantly different, despite both marks being traditionally associated with transcriptionally active chromatin. (Chapter 3). Much of this work utilizes bead-loading in order to introduce proteins and DNA into living cells. A simple, effective, and cheap procedure, bead-loading is a highly effective and versatile technique that is generally underutilized. To facilitate communication of this process, a detailed protocol is included (Chapter 4). While this culmination of work furthers our understanding of cellular genetic expression and eukaryotic transcription, it also introduces many new questions that are promising areas of study. Fortunately, the combination of imaging technology and knowledge developed here provides promising new fronts for studying transcription in living cells (Chapter 5).