UC San Diego

Genomic alterations affecting 50 or more base pairs—including amplifications, deletions, and rearrangements—are pervasive drivers of cancer evolution. Despite their importance, these large-scale somatic alterations remain poorly understood due to challenges in detecting and interpreting them from short-read sequencing data. They are broadly classified into two major categories: copy number alterations (CNAs), which involve gains or losses of DNA segments and change the number of copies of specific genomic regions, and structural variants (SVs), which rearrange the genome through events such as deletions, duplications, inversions and translocations, often disrupting gene structure or regulation. While mutational signature frameworks for single-base substitutions are well established, comparable frameworks for CNAs and SVs are still limited, hindering insights into the mutational processes shaping cancer genomes. In Chapter 1, I present a robust software implementation for systematic mutational signature analysis of large-scale alterations. The tool integrates two complementary classification frameworks—an established schema for SVs and a newly developed framework for CNAs—and supports multiple data types, input formats, and extensive visualization. It standardizes analysis and provides the first broadly accessible platform for large-scale mutational profiling. In Chapter 2, I evaluate Latent Dirichlet Allocation (LDA) as an alternative to nonnegative matrix factorization (NMF) for mutational signature extraction. LDA effectively captures mutational patterns and offers a complementary approach for discovering previously unrecognized processes driving large-scale genomic changes. In Chapter 3, I present the most comprehensive analysis to date of extrachromosomal DNA (ecDNA)—a highly amplified circular DNA structure—in lung cancers from never-smokers and smokers. Using a large, diverse cohort spanning multiple ancestries, I adapted an ecDNA detection algorithm to account for low tumor purity and variable ploidy, common in lung tumors. Genomic and epidemiological analyses revealed key ecDNA-associated features, including recurrent oncogenes, regulatory elements, chromothripsis scars, and unique mutational signatures. Notably, ecDNA was strongly linked to whole-genome doubling, specific driver mutations, and worse survival in never-smokers, emphasizing its clinical relevance. Together, this dissertation delivers essential computational tools and new biological insights, providing the most detailed characterization of ecDNA in lung cancer to date and advancing our understanding of large-scale cancer genome evolution.