Abstract

Antimicrobial Resistance (AMR) is a growing global health threat. It happens when bacteria or other pathogens evolve to resist antibiotics. The relevant infections would cause at least 50,000 deaths yearly in Europe and the United States alone1. To combat the crisis, I developed a data-driven theme for quick and accurate detection of bacterial resistance to a wide range of antibiotics. We utilized statistical approaches and developed a series of novel and rigorous models to tackle the problem of antibiotic resistance.

We proposed a deep learning-based Pan-Antibiotic Resistance Prediction model (PARP). This model enabled AMR prediction for a wide range of pathogen-antibiotics combinations. Then, we developed a Bayesian Neutral Network-based framework to explain deep learning models using rank statistics and Layer-wise Relevance Propagation. This method can elucidate the reasoning behind the prediction results of complex neural network models, demonstrated by multiple datasets, including image and genetic sequence datasets. We also proposed a Transfer Learning-based antibiotic resistance prediction framework suitable for chronic infectious diseases. Our work is novel in solving the domain shift problem between bacteria samples from different hosts by retraining the input layers in the PARP model. Altogether, our work demonstrated that statistics could enhance the development and interpretation of deep neural networks and address emerging and critical biomedical problems.

Degree Date

Spring 2023

Document Type

Dissertation

Degree Name

Ph.D.

Department

Statistical Science

Advisor

Dr. Xiaowei Zhan

Subject Area

Statistics

Format

.pdf

Creative Commons License

Creative Commons Attribution-Noncommercial 4.0 License
This work is licensed under a Creative Commons Attribution-Noncommercial 4.0 License

Available for download on Sunday, May 04, 2025

Share

COinS