Dissertation Defense: Juan Xie

Dissertation Defense: Juan Xie   

Title: Statistical Methods for Design and Analysis of Single-cell and Spatial Omics Experiments

Abstract: High-throughput spatial transcriptomics (HST) and single-cell T cell receptor sequencing (scTCR-seq) are transformative genomic technologies for studying tissue organization and adaptive immunity, respectively. While these data received significant attention from the field, HST still lacks principled guidelines for experimental design, while scTCR-seq data analysis remains challenging due to its complexity. This dissertation addresses these challenges.

1. We developed spaDesign, a novel statistical framework for HST experimental design. By combining a Poisson Gaussian process (GP) with a Fisher-Gaussian (FG) kernel mixture model to effectively simulate gene spatial expression and location patterns, spaDesign allows rigorous HST experimental design in the sense of sequencing depth determination.
    
2. We further improved computational efficiency of spaDesign with various approximation approaches, including nearest-neighbor GP and bootstrapping for GP, and an EM algorithm for the FG model. These approaches allow achieving orders-of-magnitude speedup over MCMC and facilitating interactive experimental design.
    
3. We developed LRT, a computational framework for integrative analysis of scTCR-seq and single-cell RNA-seq (scRNA-seq) data to explore clonal differentiation trajectory heterogeneity, based on a principal curve algorithm, Dirichlet-Multinomial mixture model, and permutation tests.
    

Advisor: Dr. Dongjun Chung