SE PhD Prospectus Defense of Taiyao Wang
- Starts: 11:00 am on Monday, November 5, 2018
- Ends: 1:00 pm on Monday, November 5, 2018
ABSTRACT: Analytics and optimization theory are well-developed techniques to describe, predict and optimize real-world systems and they have been widely used in engineering and science. In this prospectus, we focus on applications in biomedical systems ranging from microbial communities to the human body.
In the first problem, we consider predictions of 30-day hospital readmissions. We apply standard algorithms and develop novel interpretable learning algorithms that can reliably predict readmissions within 30-days of discharge of patients undergoing general surgery.
In the second problem, we present a machine learning framework and a data-driven approach for supporting clinical decision systems. For this problem, we explore and develop predictive and prescriptive analytics under the Support Vector Machines (SVM) framework. The predictive and prescriptive analytics are shown to predict and prevent 30-day hospital readmissions after general surgery using a dataset available from the National Surgical Quality Improvement Program (NSQIP) which contains over 2.28 million de-identified patients from 2011 to 2014.
In the last problem in this prospectus, we formulate a novel problem to design metabolic division of labor in microbial communities. We consider a given number of microbial species living in a community and a list of all metabolic reactions present in the community, expressed in terms of the metabolite proportions involved in each reaction. We leverage tools from Flux Balance Analysis (FBA) and formulate the problem as a mixed integer linear programming problem. The strategies we find reveal a large space of nuanced and nonintuitive metabolic division of labor opportunities, including, for example, splitting the TCA cycle into two separate halves. More broadly, we systematically mapped the landscape of possible 1-, 2-, and 3-strain solutions at increasingly tight constraints on the number of allowed reactions.
Our ongoing work can be summarized as follows:
We work in collaboration with the Boston University School of Public Health and Boston Medical Center to develop prediction models for predicting female pregnancy using a dataset (PRESTO) obtained from surveys of female participants. Using electric health record data, we will develop a predictive model to predict the presence of an ovary condition, Polycystic ovary syndrome (PCOS), which is a leading cause of infertility. Finally, we will also develop a joint framework for clustering and regression.
COMMITTEE: Committee: Advisor: Yannis Paschalidis, SE/ECE; Daniel Segrè, Biology/BME/Bioinformatics; Christos Cassandras, SE/ECE; Pirooz Vakili, SE/
- 15 Saint Mary's Street, Room 121