Hepatobiliary cancers digital twins: Integrating artificial intelligence-driven multi-omics with organoid dynamics

Static biomarkers are increasingly failing to predict outcomes in hepatobiliary cancers, largely due to the profound heterogeneity of the tumor microenvironment. Over the past decade, rapid advances in digital pathology, single-cell profiling, and spatial multi-omics have generated increasingly detailed maps of tumor architecture. Yet, these static snapshots fail to capture the dynamic cellular interactions driving disease progression. Conversely, patient-derived organoids (PDOs) and organ-on-chip models capture functional drug responses but often lack the microenvironmental complexity of the original tumor. In this review, we use the term digital twin to refer to a patient-specific model that is initialized with individual baseline data, linked to measurable biological or clinical readouts, and refined with longitudinal observations. We outline a unified three-layer architecture: A Data Layer (digitizing static atlases), a Physical Layer (engineering high-fidelity PDO avatars using tunable biomaterials), and a Computational Layer (integrating PDOs and patient data). Moreover, we discuss clinical applications, including individualized therapy selection, exploration of resistance mechanisms, and optimization of drug combinations and treatment sequences. Finally, we highlight the major barriers to clinical translation, including limited paired datasets, variable PDO feasibility, and the difficulty of integrating data collected at different time scales. Taken together, these advances may support more individualized and adaptive care in hepatobiliary cancers.