From Automated Pipelines to Spatial Deconvolution
Open Access DepositedAdvanced Computational Methods for Single-Cell and Spatial Transcriptomics Analysis
Downloadable Content
a mouse skin injury model revealing seven distinct fibroblast subtypes critical for wound healing and a human skin aging dataset demonstrating age-related loss of fibroblast functional specialization. The automated pipeline successfully processes datasets ranging from thousands to millions of cells while maintaining statistical rigor and biological interpretability, significantly reducing computational overhead and enhancing reproducibility. Chapter 2 advances our understanding of intercellular communication by developing a Graph Attention Network (GAT) framework for predicting ligand-receptor interactions. Unlike conventional database-driven approaches, our deep learning architecture learns complex spatial and molecular patterns from high-dimensional single-cell data, enabling the discovery of context-specific signaling pathways. The GAT model integrates gene expression profiles with network topology, capturing both direct receptor-ligand binding and downstream signaling cascades. Through systematic benchmarking across multiple tissue types, we demonstrate superior performance in identifying biologically relevant interactions, particularly in detecting tissue-specific and condition-dependent communication networks that traditional methods fail to capture. In Chapter 3, we apply cellSight to comprehensively characterize transcriptional differences across diverse human skin types, investigating how genetic background and environmental factors shape cellular composition and gene expression patterns. By analyzing single-cell profiles from donors representing different ancestries and phototypes, we identify skin type-specific gene signatures in keratinocytes, melanocytes, and fibroblasts. Our differential expression analysis reveals distinct regulatory programs governing pigmentation, barrier function, and inflammatory responses across skin types. Furthermore, cell-cell communication analysis uncovers differential signaling networks between melanocytes and keratinocytes that contribute to skin type-specific phenotypes. These findings provide molecular insights into skin diversity and have important implications for personalized dermatology and the development of skin type-appropriate therapeutics. Chapter 4 addresses a critical gap in spatial transcriptomics analysis by developing a novel deconvolution method that leverages scRNA-seq reference data to estimate cell type composition in spatial transcriptomics spots. Traditional deconvolution approaches often struggle with the unique characteristics of spatial data, including spot-level mixtures of multiple cell types and spatial dependencies between neighboring regions. Our method integrates probabilistic modeling with graph-based regularization, incorporating both molecular signatures from scRNA-seq references and spatial neighborhood information from tissue architecture. We employ a constrained optimization framework that ensures biologically plausible cell type proportions while preserving spatial continuity. Validation on well-characterized tissues, including mouse brain and human skin, demonstrates that our approach accurately reconstructs cellular composition with significantly improved spatial coherence compared to existing methods. The framework enables high-resolution mapping of cell type distributions, revealing spatial organization patterns and microenvironmental niches that drive tissue function. This dissertation presents an integrated computational ecosystem for single-cell and spatial transcriptomics analysis. From cellSight’s automated preprocessing to advanced GAT-based communication networks and spatial deconvolution, these methods establish a comprehensive framework for extracting biological insights from complex genomic datasets. The open-source implementation of these tools democratizes access to sophisticated analytical capabilities, accelerating discovery in developmental biology, disease pathogenesis, and personalized medicine. Our work on skin biology demonstrates the power of these approaches in revealing cell type diversity, intercellular communication networks, and spatial organization principles that govern tissue homeostasis and disease. By bridging the gap between methodological innovation and biological application, this research contributes to the growing field of computational biology and provides practical tools that will enable researchers to unlock the full potential of single-cell and spatial transcriptomics technologies.
The emergence of single-cell RNA sequencing (scRNA-seq) and spatial transcriptomics has revolutionized our understanding of cellular heterogeneity and tissue architecture. However, the computational complexity of analyzing these high-dimensional datasets presents significant challenges, including manual preprocessing, parameter optimization, and the integration of spatial information with molecular profiles. This dissertation addresses these challenges through the development of innovative computational frameworks that span automated analysis pipelines, advanced network-based methods for intercellular communication, and novel approaches for spatial data deconvolution.In Chapter 1, we introduce cellSight, a comprehensive automated workflow for single-cell RNA sequencing analysis that integrates quality control, normalization, dimensionality reduction, differential expression analysis using Tweedieverse, and cell-cell communication inference through CellChat. We validate cellSight on two distinct biological systems
- All rights reserved
Notice to Authors
If you are the author of this work and you have any questions about the information on this page, please use the Contact form to get in touch with us.