Computational Foundations of Tumor Evolution and Intratumor Heterogeneity Inference

Cancer progression is fundamentally an evolutionary process: subpopulations of cells accumulate distinct mutations and are shaped by selection, giving rise to extensive intratumor heterogeneity (ITH) that drives treatment resistance and relapse. Reconstructing this evolutionary history from sequencing data is a rich source of computational problems, blending combinatorial optimization, probabilistic modeling, and statistical learning. This talk will give an overview of the computational foundations of tumor evolution, with the aim of three key questions. First, how can mixed sequencing signals, from bulk sequenced tissue or low-coverage single cells, be deconvolved into distinct clonal populations? Second, how can phylogenetic trees relating these clonal cell populations be reconstructed, subject to biologically motivated constraints? Third, how can discordant evolutionary signals across multiple data modalities, including single-nucleotide and structural alterations, be reconciled into a single consistent model of tumor history? The talk will present algorithmic methods to address these questions and will discuss how they extend to emerging single-cell and long-read sequencing technologies, where the volume and structure of the data open new opportunities and new computational bottlenecks. It will also touch on the statistical challenges of uncertainty in this setting, where ground truth is rarely available and model misspecification is the norm rather than the exception. Throughout, the talk will highlight open problems at the interface of combinatorics, statistics, and cancer biology, making the case that tumor evolution is not just an application domain for existing techniques but a source of new computational questions.