Genome instability (GI) is a hallmark cancer that drives both tumorigenesis and evolution. Despite its ubiquitous nature, we have limited understanding of GI’s origins, heterogeneity, and clinical implications. Mutations underlying GI arise from an admixture of exogenous and endogenous factors – each generating specific lesions across mutational modalities (e.g. single base substitutions [SBS], structural variations [SVs], etc.). The relative contributions of these factors can vary temporally and by tissue. This high combinatorial complexity makes comprehensive identification of distinct GI patterns intractable experimentally. Instead, the presence, activities, and sources of GI patterns can be derived computationally – particularly from vast collections of cancer genomes. Doing so highlights dysregulated mutational processes throughout a malignancy’s life history. Disentangling these processes and their interactions provides critical information on how cancers develop, can be treated, and can be prevented.

Within the Pitt Lab, we leverage big data genomics and artificial intelligence (AI) to identify GI patterns pan-cancer along with their causes, consequences, and context-dependencies. We approach this problem in multiple unique ways: i) by deriving multimodal GI patterns through innovative learning methods; ii) by developing generative AI models to predict these patterns under specific conditions (e.g. tissue, DNA repair deficiencies, etc.); and iii) with experimental validation of these predictions through synergistic collaborations. Together, these approaches address three thematic questions:

1. What are the multimodal GI patterns that emerge on cancer genomes?
2. What do these patterns tell us about cancer’s evolution and heterogeneity?
3. Can these patterns guide cancer treatment and prevention?