Allele-specific somatic copy number alteration analysis in single-cell or spatial transcriptomics data

Abstract

Somatic copy number alterations (CNAs) are critical genomic mutations that contribute to the onset and progression of diverse cancer types. Characterization of somatic CNAs with resolution at the individual cell and gene level is crucial for comprehensive analyses of cancer evolution and the delineation of clonal populations. High-throughput single-cell RNA sequencing (scRNA-seq) provides a powerful technology to dissect the genetic underpinnings of tumor heterogeneity in a single assay. This fundamental technique has fueled the development of many computational strategies that employ statistical models to analyze read counts for the detection of somatic CNAs. Nonetheless, the inherent technical limitations of sparsity in such single-cell transcriptomic data pose obstacles to the accurate detection of allele-specific somatic CNAs, especially within intricate clonal architectures. This thesis contributes to the development of statistical methods for detecting somatic CNV by integrating phased allelic information with expression signals to achieve more accurate and robust inference across multiple cancer types using scRNA-seq datasets. First, we developed a statistical framework, XClone, that strengthens the signals of read depth and allelic imbalance through effective smoothing across cellular neighborhoods and gene coordinate graphs. This approach facilitates the identification of haplotype-aware CNAs within scRNA-seq data at single-cell and single-gene resolution. By applying XClone to multiple datasets with challenging complex compositions, we demonstrated its ability to robustly detect different types of allele-specific CNAs and potentially indicate whole genome duplication, therefore enabling the discovery of corresponding subclones and the dissection of their phenotypic impacts. Second, we performed comprehensive benchmarking research on five recent widely-used single-cell RNA sequencing CNV detection methods (InferCNV, CopyKAT, CaSpER, Numbat, and XClone) across 15 distinct scenarios, including one scRNA-seq dataset with a solid omics-based ground truth and 14 simulated tasks that mimic a spectrum of challenging data conditions. Through simulations of allelic CNAs, XClone's robust performance was confirmed, and the collated benchmarking results offer a comprehensive guide, assisting users in method selection for their specific data analyses. Third, we extended XClone to adapt to the spatial transcriptomics data and applied it to a prostate cancer specimen. Utilizing precise histological annotations for validation, we discovered the sensitivity of allelic signals to variations within the tumor microenvironment (TME), thereby emphasizing the crucial role of B allele frequency (BAF) in the spatial transcriptomic analysis of tumor samples. Additionally, We evaluated several methods, including XClone, using a 10X Visium dataset from a prostate sample to determine their effectiveness with spatial transcriptomics data and their cross-platform applicability. In summary, this thesis proposes a structured statistical framework (XClone) that significantly refines the discernment of allele-specific somatic CNAs at an unprecedented resolution of single cell and single gene.