From Cancer Initiation to Clinical Insight: Computational and Machine Learning Approaches to Tumor Dynamics and Precision Oncology

Abstract

Despite extensive research, many fundamental questions in cancer biology and clinical oncology remain unanswered, from the mechanisms underlying cancer initiation and early development to factors that influence diagnosis, treatment response, and patient outcomes. Computational modeling and machine learning frameworks provide powerful tools to address these challenges by enabling large scale mechanistic modeling and the integration of complex molecular and clinical information to uncover mechanisms, generate hypotheses, and guide clinical decisions. This thesis combines computational modeling of cancer initiation and machine learning frameworks for precision oncology to bridge fundamental and translational aspects of cancer research. Studies of cancer initiation are impeded by complications of identifying and tracking the cell of origin. Recent work has shown that mutagen-induced DNA lesions can persist over multiple rounds of cell division, leaving a statistically interpretable footprint of cancer initiating events. Specifically, it allows estimation of the number of divisions between the DNA lesion introduction and the most recent common ancestor of the developed tumor (LAD). We developed a branching process model of cancer evolution following lesion introduction, and analyzed footprints of segregating lesions from previously published experimental mouse data and post-chemotherapy human metastatic tumors to obtain LAD estimates. We show that in all contexts cancer clones tended to start early, usually within 4 cell generations. Analytical and computational implementations of the branching process model suggested the fitness advantage of early cancer drivers must have exceeded 30% to achieve such early clone initiation. At the clinical end of the spectrum, precision oncology has informed cancer care by enabling the discovery and application of diagnostic, prognostic, and/or predictive molecular biomarkers. However, many patients lack actionable biomarkers or fail to respond to biomarker-directed therapies. Patient similarity approaches can leverage comprehensive tumor profiling and prior clinical experiences from large cohorts for decision support, facilitating broader realization of precision oncology benefits. We developed a deep learning based modeling framework using real-world clinicogenomic data from a tertiary cancer center to (i) measure patient similarity based on embedded tumor genomic profiles and (ii) evaluate the association of derived patient subgroups and neighborhoods with shared therapeutic outcomes in a breast cancer specific and in a histology-agnostic pan-cancer setting. The model recovered clinically meaningful patient groups of both expected and previously unknown therapeutic associations, as well as patient-specific neighborhoods that could inform therapeutic trajectories more often than expected by chance in multiple clinical contexts. Moreover, model utility extended to patients without actionable genomic biomarkers and those with cancer of unknown primary (CUP) diagnoses, where neighborhoods aligned with independently predicted primary cancer type. These neighborhoods could also be examined over time in a continuously learning scenario. Overall, these studies integrate fundamental models of cancer evolution and translational machine learning to develop approaches that advance cancer research from onset to clinical decision making. Our branching process model allowed inference of tumor initiation and growth parameters based on events preceding the most recent common ancestor of the initiating clone as opposed to characteristics of fully grown tumors. In parallel, our similarity-based modeling framework distilled complex molecular and clinical data into concise, context-specific insights that augment rather than replace clinician judgment, providing a foundation for real-time learning, patient-centered decision support in precision oncology.