Abstract
Large-scale genomic and functional datasets have transformed the study of cancer, but realizing their full potential requires analytical frameworks that go beyond single-gene, single-variable thinking. Systems biology approaches including integrative multivariate modeling, information-theoretic feature selection, and dimensionality reduction offer a means of interrogating coordinated molecular patterns that conventional analyses cannot detect. In melanoma, a cancer with well-defined oncogenic drivers but incompletely explained clinical and biological heterogeneity, these approaches can reveal regulatory dependencies and genomic architectures that driver-centric classifications miss. This dissertation applies complementary computational frameworks to two under-characterized dimensions of melanoma heterogeneity: the epigenetic complex dependencies that sustain malignant transcriptional states, and the mutational architecture underlying in-transit melanoma, a clinically distinct pattern of metastatic dissemination.
Epigenetic regulation plays a fundamental role in maintaining the gene expression programs that define cellular identity and support tumor cell survival. In cancer, dysregulation of these mechanisms creates context-specific dependencies on chromatin regulators that can represent therapeutic vulnerabilities. Because epigenetic regulators function within multi-subunit complexes rather than as isolated enzymes, these dependencies likely arise from coordinated complex activity and may not be apparent through single-gene analyses. In the first project, I developed a complex-level analytical framework integrating genome-wide CRISPR dependency data from over 1,000 cancer cell lines with curated chromatin complex annotations and applied a lineage-aware statistical modeling approach to evaluate epigenetic dependencies across 42 cancer lineages. The analysis revealed structured organization of chromatin dependencies across lineages and identified the H3K4 methyltransferase complex Set1C/COMPASS as a previously unrecognized melanoma-enriched vulnerability. A multivariate analysis across 735 epigenetic regulators showed that this dependency is not restricted to a specific differentiation state, distinguishing it from chromatin vulnerabilities that vary along the melanocytic-to-undifferentiated transcriptional spectrum. Functional validation confirmed that Set1C/COMPASS activity is required to maintain global H3K4 trimethylation and proliferation, with CXXC1 depletion inducing cell-cycle arrest and suppressing MYC- and E2F-driven transcriptional programs.
In parallel, certain melanoma presentations remain poorly characterized at the molecular level despite their clinical relevance. In-transit melanoma, a form of regional metastatic disease defined by intralymphatic tumor deposits arising between the primary lesion and the regional lymph node basin, occurs in approximately 5-10% of patients, is associated with high locoregional recurrence, and lacks recognized molecular markers. In the second project, I applied an information-theoretic framework to targeted sequencing data from the MSK-IMPACT cohort to compare the mutational profiles of in-transit melanoma lesions with those of primary tumors, regional lymph node metastases, and distant metastases. Extending mutual information analysis to pairwise gene combinations that incorporated both mutant and wild-type states identified a reproducible mutational pattern centered on NRAS Q61 mutations. Among the strongest associations were the retained wild-type status of several regulatory genes, most notably PAK5 and PRKN. Expanding this analysis to include the broader set of genes that consistently remained wild-type in the presence of NRAS mutations revealed a larger combinatorial genotype. Functional annotation of these genes pointed to signaling pathways including PI3K/AKT/mTOR, TGF-β, and E2F-associated programs as potential features of this mutational context. Importantly, these combinatorial patterns proved more discriminatory than any single alteration alone and were independently validated in a separate patient cohort.
Together, these studies show that meaningful dimensions of melanoma heterogeneity emerge from coordinated regulatory dependencies and combinatorial mutational architectures that gene-by-gene approaches do not capture. The frameworks developed here offer broadly applicable strategies for investigating higher-order molecular features in cancer, both in contexts where known drivers insufficiently explain biological diversity, and in contexts where systematic molecular characterization has been largely absent.