In partial fulfillment of the requirements for the degree of

Doctor of Philosophy in Bioinformatics

in the Department of for Biomedical Engineering

Xinling Li

Will defend her thesis:

Quantitative Approaches for Identifying the Limitation of Single-cell Sequencing Technologies and Applying Sequencing Technologies in the Study of Infectious Diseases and Cancer

May 7, 2026

10:00 AM – 12:00 PM (EST)

EBB 4029

https://teams.microsoft.com/meet/23036548151103?p=UplOBRjEVYTj2NYCAm

Thesis Advisor:

Dr. Qiu Peng

Department of for Biomedical Engineering

Georgia Institute of Technology

Committee Members:

Dr. Gabe Kwong

Department of for Biomedical Engineering

Georgia Institute of Technology

Dr. Greg Gibson

School of Biological Sciences

Georgia Institute of Technology

Dr. May Wang

Department of for Biomedical Engineering

Georgia Institute of Technology

Dr. M.G. Finn

School of Chemistry and Biochemistry

Georgia Institute of Technology                

Abstract:

This thesis focuses on identification of the limitation of single-cell sequencing technologies, and development of pipeline and models for analyzing sequencing data. The aim of this work is to contribute to a deeper understanding of sequencing technologies and a better utilization of the technologies in solving challenges in infectious diseases and cancer research. We first identified systematic under-detection of genes in scRNA-seq and Visium data through analyses of paired bulk RNA-seq and single-cell genomics datasets. These genes were enriched for poly(T) motifs near their 3′ ends, leading to the hypothesis that such motifs may form hairpin structures with poly(A) tails of mRNA transcripts, reducing capture efficiency during library preparation and explaining their lower detection compared to bulk RNA-seq. We also proposed a framework integrating graphical models and conventional machine learning to leverage scRNA-seq data for predicting COVID-19 severity and identifying biomarkers. Graphical models effectively distinguished healthy individuals from infected patients (F1-score > 0.9), while traditional machine learning accurately classified disease severity (F1-score > 0.99), revealing meaningful molecular signatures. In addition, we developed a bioinformatics pipeline to predict existence of proteases across bacterial species by leveraging known bacteria–protease relationships. Through large-scale protein–genome alignments, we identified previously unreported proteases across bacterial genomes that cause severe infectious diseases, with validation from bulk RNA-seq data. Finally, we constructed mathematical models based on mass action kinetics to evaluate multi-arm protease sensors, demonstrating their potential advantages in cancer detection. Overall, this work highlights limitations of current sequencing technologies and provides computational solutions to enhance their utility in early cancer detection and infectious disease treatment and diagnostics.