Presented By: Michigan Institute for Computational Discovery and Engineering
Chanese Forte (Environmental Health Sciences) and Hyeon Joo (Health Infrastructure and Learning Systems)
Ph.D. in Scientific Computing Student Seminar Series
CHANESE FORTE: Chanese is a Dual PhD student pursuing a degree in the Environmental Health Sciences and Scientific Computing. Chanese’s research interests lie in chemical exposure in agriculture workers and cellular alteration.
"ASCERTAINING PESTICIDE EXPOSURE AND BIOACTIVITY USING OPEN SOURCE DATA: Pesticides are known to be harmful chemicals to human health, however they are still heavily used in agriculture. Using 3 large publicly available datasets, this study aims to 1) quantify pesticide exposure levels of the US general population in comparison to farmworkers, and 2) assess biological activity of these pesticides using ToxCast high throughput screening toxicity data. The National Health and Nutrition Examination Survey (NHANES) is a cross-sectional study representative of the US population with oversampling weights for minoritized populations. NHANES was used to quantify pesticide exposure among US farmworkers and the general population who responded to NHANES between 1999 and 2016. Overall, there are 65,893 farmworkers in the NHANES dataset with at least one pesticide biomarker present in the laboratory, and a total of 92,140 people total in NHANES. Biomarker measurements were available for 28 different pesticides, which also have been assessed for dose-dependent toxicity using high throughput assays in ToxCast. Of the 28 pesticides, Mirex had the lowest average exposure concentration at 1.19×10- 11mol/L among US farmworkers, whereas p,p’-DDE had the highest average exposure level at 5.4×10- 7mol/L. In ToxCast, the AC50 represents the concentration at 50% of the maximum biological activity. The average AC50 for p,p’-DDE is 1.3mol/L suggesting that activation levels are considerably higher than that of the average farmworkers’ exposure levels. It compares and analyzes through regression, the US pesticide exposure levels to the bioactivity of these same pesticides within the human body. By comparing population level data with toxicological assay data, we hope to create a more overarching idea of how pesticides may be affecting the body and a human population level.
HYEON JOO: Hyeon Joo is a second year PhD student in Health Infrastructures and Learning Systems. He completed his MS in Computer Science and Engineering, and Master of Health Informatics from the University of Michigan, Ann Arbor. His research focuses on developing and implementing computational data-driven algorithms, systems or tools to help users identify gaps and make informed decisions. He loves working in the field of health care as a data scientist and a software engineer.
"EARLY PREDICTION OF HEART FAILURE USING ATTENTION MODELS USING EHR DATA": Heart Failure (HF) is a severe and progressive chronic condition affecting over 5.8 million patients with a 5-year mortality rate of 45-60% in the United States. Despite significant efforts and advanced HF management, diagnosing HF in the early stages remains challenging due to its syndromic nature and non-specific disease presentation. In this seminar, I will present a single attention recurrent network and a hierarchical attention convolutional neural networks to detect the early stage of HF at a tertiary hospital. I will also describe various methods of feature selection to reduce the computation time and improve the performance of the models. Lastly, I will present the challenges of adopting models in clinical practice which leads to my next research steps.
Register to receive Zoom login information: https://myumi.ch/BobQB
"ASCERTAINING PESTICIDE EXPOSURE AND BIOACTIVITY USING OPEN SOURCE DATA: Pesticides are known to be harmful chemicals to human health, however they are still heavily used in agriculture. Using 3 large publicly available datasets, this study aims to 1) quantify pesticide exposure levels of the US general population in comparison to farmworkers, and 2) assess biological activity of these pesticides using ToxCast high throughput screening toxicity data. The National Health and Nutrition Examination Survey (NHANES) is a cross-sectional study representative of the US population with oversampling weights for minoritized populations. NHANES was used to quantify pesticide exposure among US farmworkers and the general population who responded to NHANES between 1999 and 2016. Overall, there are 65,893 farmworkers in the NHANES dataset with at least one pesticide biomarker present in the laboratory, and a total of 92,140 people total in NHANES. Biomarker measurements were available for 28 different pesticides, which also have been assessed for dose-dependent toxicity using high throughput assays in ToxCast. Of the 28 pesticides, Mirex had the lowest average exposure concentration at 1.19×10- 11mol/L among US farmworkers, whereas p,p’-DDE had the highest average exposure level at 5.4×10- 7mol/L. In ToxCast, the AC50 represents the concentration at 50% of the maximum biological activity. The average AC50 for p,p’-DDE is 1.3mol/L suggesting that activation levels are considerably higher than that of the average farmworkers’ exposure levels. It compares and analyzes through regression, the US pesticide exposure levels to the bioactivity of these same pesticides within the human body. By comparing population level data with toxicological assay data, we hope to create a more overarching idea of how pesticides may be affecting the body and a human population level.
HYEON JOO: Hyeon Joo is a second year PhD student in Health Infrastructures and Learning Systems. He completed his MS in Computer Science and Engineering, and Master of Health Informatics from the University of Michigan, Ann Arbor. His research focuses on developing and implementing computational data-driven algorithms, systems or tools to help users identify gaps and make informed decisions. He loves working in the field of health care as a data scientist and a software engineer.
"EARLY PREDICTION OF HEART FAILURE USING ATTENTION MODELS USING EHR DATA": Heart Failure (HF) is a severe and progressive chronic condition affecting over 5.8 million patients with a 5-year mortality rate of 45-60% in the United States. Despite significant efforts and advanced HF management, diagnosing HF in the early stages remains challenging due to its syndromic nature and non-specific disease presentation. In this seminar, I will present a single attention recurrent network and a hierarchical attention convolutional neural networks to detect the early stage of HF at a tertiary hospital. I will also describe various methods of feature selection to reduce the computation time and improve the performance of the models. Lastly, I will present the challenges of adopting models in clinical practice which leads to my next research steps.
Register to receive Zoom login information: https://myumi.ch/BobQB
Related Links
Co-Sponsored By
Explore Similar Events
-
Loading Similar Events...
