PROJECT SUMMARY/ABSTRACT Cancer progression is an evolutionary process, enabling tumors to evade our best therapeutics and, ultimately, to recur and metastasize. The evolution of therapeutic resistance remains enigmatic in part because it must be reconstructed from patient biopsies which are prescribed by patient care, rather than from design to empower discovery. Some mechanisms of resistance in select cancer-therapy systems have been discovered, but this knowledge is the result of extensive work and is subject to obsolescence as therapeutic strategies themselves evolve, necessitating systematic approaches to reveal how therapy is subverted. Our goal is to leverage tumor phylogenies to examine the evolutionary forces operating on variants responsible for therapeutic resistance. To expose the timing of specific SNVs and quantify their contribution to therapeutic resistance, I will perform deconvolution of selection from the underlying mutation rate on therapy-exposed branches of tumor phylogenies to obtain cancer effect sizes for SNVs in the context of therapy. I employed this approach in a preliminary analysis of EGFR L850R-driven, erlotinib-treated lung adenocarcinoma, and it revealed that the EGFR T790M mutation in has a vast effect size—roughly 45× higher in the context of therapy than the L858R mutation driving primary progression has during primary progression. The results support molecular and clinical observations that EGFR T790M confers strong erlotinib resistance and suggest vulnerability of EGFR L850R-driven cancer lineages to simultaneous, orthogonal treatments. Application of this method to less well characterized cancer-therapeutic systems will enable detection of currently unknown enablers of resistance. Development of these approaches to illuminate the evolutionary dynamics of therapeutic resistance will reveal the series and timing of mutations enabling resistance. To fully utilize this knowledge, we will perform analyses of somatic variant epistasis on cancer chronograms, resolving the gene interactions and evolutionary trajectories that lead to therapeutic resistance. Further, we will learn from convergences in the mechanisms of therapeutic resistance, modifying a Markov Chain Monte Carlo approach to the Bayesian reconstruction of cancer chronograms, to quantify—on average—the time until specific mutations reach high frequency. Finally, we will develop synergistic phylogenetic and machine-learning strategies that characterize the etiology of unobservable ancestral features of cancer. For instance, our initial analysis of a lung adenocarcinoma patient receiving erlotinib preceded by cisplatin found that platin-induced mutational signature peaked on the same branch of the phylogeny as the EGFR c.2369C→T substitution causing the T790M resistance mutation, suggesting that preceding erlotinib therapy with cisplatin can supply exactly the genetic heterogeneity needed by the tumor to evade erlotinib therapy. These approaches will also...