Integration of Multi-Omics in AI-Based Models for Predictive Oncology

Cancer heterogeneity and the emergence of innovative therapeutic alternatives have made the development of tools to help in clinical decision making even more necessary. The recent advances in the field of artificial-intelligence-based predictive models make them promising allies due to their ability to integrate vast amounts of descriptors and data, notably omic data, for grasping cancer complexity. However, it is difficult to present a complete picture of this field due to the increasing number of models and the many possible applications. The aim of this review was to offer an overview of omic-based predictive models in oncology through the prism of three important concepts: the objective, the data, and the model. We show that the consideration of these three aspects could help to assess the feasibility, validity and usability of a model and to guide its construction. Lastly, we highlight that taking into account the underlying technical, physiological and biological phenomena inherent to omic data at each step of the project may improve the model predictive capability and its explainability.