Integrating Multimodal and Temporal Patient Representations for Cancer Diagnosis and In Silico Drug Development

Background:Oncology is increasingly a data-rich discipline, yet clinical and research data remain fragmented across modalities. A single patient may be characterized by hematopathology, histopathology and radiology images, genomic and transcriptomic profiles, and longitudinal clinical records, but these streams are typically analyzed in isolation. This fragmentation limits diagnostic accuracy in complex and rare malignancies and constrains the biological insight available for therapeutic development. Recent advances in foundation models and representation learning offer a path to integration, rather than task-specific pipelines, a single model can learn compact, structured embeddings that unify heterogeneous inputs. Extending such representations to capture temporal dynamics, how a patient's state evolves, remains largely unaddressed. A unified, multimodal, and temporal patient representation could therefore bridge diagnosis, biomarker discovery, and in silico modeling within a unified framework.Research Goal:This Research Topic aims to advance biological foundation models that learn unified patient representations with compact, structured embeddings integrating imaging, molecular, and temporal clinical data into a single latent representation per patient. We seek contributions that treat integration not only as a downstream fusion step but also as the central modeling objective. The representations are intended to serve three coupled aims. First, earlier and more reliable diagnosis of complex and rare cancers, with a particular focus on leukemia and lymphoma. Second, the discovery of novel molecular and digital biomarkers that emerge from cross-modal structure. Third, exploratory virtual patient simulation to inform and de-risk drug development, where high R&D costs and late-stage clinical trial failure remain a central bottleneck. By connecting these aims within one representational framework, the Topic seeks methods that are simultaneously clinically actionable and biologically interpretable, bridging computational innovation and translational medicine.Scope and Information for authors:The research topic welcomes original research, methods, review, and perspective articles addressing unified patient representations in pathology and hematology. We encourage contributions on multimodal integration of imaging, molecular, and clinical data; foundation models and self-supervised representation learning for pathology, hematology, radiology, and omics; temporal and longitudinal modeling of disease trajectories; applications to diagnosis and subtyping of leukemia, lymphoma, and other complex or rare cancers; discovery and validation of molecular and digital biomarkers; and in silico or virtual patient simulation approaches supporting in silico drug development. Studies emphasizing clinical validation, interpretability, generalization across cohorts, or data-efficient learning, virtual cells world models for rare diseases are particularly encouraged. We invite Original Research, Methods, Systematic Reviews, Reviews, Mini Reviews, and Perspective articles. Submissions demonstrating translational relevance, linking computational advances to diagnostic, prognostic, or therapeutic outcomes, are strongly welcomed. Clinical computational collaborations and multicenter validation studies are also welcomed. Topic editors: Rao Muhammad Umer, Gian Luca Foresti, Hussain Ahmad Madni, Asad Munir Source: https://www.frontiersin.org/research-topics/83828/integrating-multimodal-and-temporal-patient-representations-for-cancer-diagnosis-and-in-silico-drug-development