Approaches to personalized diagnosis and treatment in oncology are heavily reliant on computer models that use molecular and clinical features to characterize an individual patient’s disease. Most of these models use genome and/or gene expression sequences to develop classifiers of a patient’s tumor. However, in order to fully model the behavior and therapy response of a tumor, dynamic models are desirable that can act like a Digital Twin of the cancer patient allowing prognostic and predictive simulations of disease progression, therapy responses and development of resistance. We are constructing Digital Twins of cancer patients in order to perform dynamic and predictive simulations that improve patient stratification and facilitate the design of individualized therapeutic strategies. Using a hybrid approach that combines artificial intelligence / machine learning with dynamic mechanistic modelling we are developing a computational framework for generating Digital Twins. This framework can integrate different types of data (multiomics, clinical, and existing knowledge) and produces personalized computational models of a patient’s tumor. The computational models are validated and refined by experimental work and in retrospective patient studies. We present some of the results of the dynamic Digital Twins simulations in neuroblastoma. They include (i) identification on non-MYCN amplified high risk patients; (ii) prediction of individual patients’ responses to chemotherapy; and (iii) identification of new drug targets for personalized therapy. Digital Twin models allow the dynamic and mechanistic simulation of disease progression and therapy response. They are useful for the stratification of patients and the design of personalized therapies.