Computational Analysis and Statistical Methods in Spatial Transcriptomics and Neuroimaging

Abstract

This dissertation delves into challenges and innovations in spatial transcriptomics and neuroimaging. Spatial transcriptomics has transformed transcriptome analysis by incorporating tissue spatial architecture. While many studies focus on spatially variable genes and clustering in spatial transcriptomics data, how different feature sets, particularly those related to spatial information, influence cell type identification remains under-explored. Similarly, analyzing functional brain connectivity networks derived from neuroimaging data is crucial for understanding brain function and mental health disorders. However, existing methods often face issues with unclear clinical interpretations or high computational costs. This dissertation addresses these challenges. Chapter 2 focuses on spatial transcriptomics. Various methods have been developed for detecting spatially variable genes (SV genes), whose gene expression exhibits strong spatial autocorrelation across the tissue space. These genes are often used to define clusters in cells or spots. However, highly variable (HV) genes, characterized by significant expression variation from cell to cell, are conventionally used in clustering analyses. In this chapter, we investigate whether adding highly variable genes to spatially variable genes can improve the clustering performance in spatial transcriptomics data. We tested the clustering performance of HV genes, SV genes, and the union of both gene sets (concatenation) on over 50 real spatial transcriptomics datasets across multiple platforms, including Vizgen’s merFISH, Nanostring’s cosMx, and 10X Genomics’ Xenium and Visium. Our results show that combining HV genes and SV genes can improve overall cell-type clustering performance. Chapter 3 introduces BSNMani, a novel Bayesian scalar-on-network regression model with manifold learning. BSNMani includes two components: a network manifold learning model for brain connectivity networks, which extracts shared and subject-specific structures, and a joint predictive model examining the link between clinical outcomes and network features while controlling for confounders. We developed a two-stage hybrid posterior computation algorithm combining the Metropolis-Adjusted Langevin Algorithm (MALA) and Gibbs sampling. BSNMani excels in extracting features from brain connectivity networks and predicting clinical outcomes, outperforming existing methods in simulations. Applied to resting-state fMRI data from the Predictors of Remission in Depression to Individual and Combined Treatments (PReDICT) study with 130 patients, BSNMani reveals connectivity among functional modules like the Somatomotor, Dorsal Attention, Salience/Ventral Attention, and Control modules related to Major Depressive Disorder (MDD) remission. This clarifies the relationship between brain connectivity and clinical attributes, enriching our understanding of psychiatric and neurological conditions. Chapter 4 develops BSNMani+, refining and scaling BSNMani for large-scale connectivity networks and clinical data. To address computational challenges from orthogonality constraints, we propose a novel prior model approximating the uniform distribution on the Stiefel manifold, leading to an efficient Gibbs sampling algorithm. BSNMani+ reduces runtime by over 50% compared to BSNMani and achieves superior parameter estimation and predictive performance. Applied to resting-state fMRI data from the Adolescent Brain Cognitive Development (ABCD) study with over 4,500 participants, BSNMani+ reveals connectivity among functional modules like the Default Mode, Visual, Fronto-Parietal, Cingulo-Opercular, and Somatomotor modules, highlighting interactions in developing adolescents. BSNMani+ was also used to analyze weighted gene co-expression networks from spatial transcriptomics and clinical data in the Seattle Alzheimer’s Disease Brain Cell Atlas (SEA-AD) study with 27 donors. Findings reveal gene co-expression modules related to neurotransmission, neural signaling, memory, and neuroprotection, enhancing our understanding of Alzheimer’s Disease pathology.