Dcode is being developed as a context-aware, multi-modal machine learning model capable of modeling gene function and cancer pathway disruption across diverse omicsepigenomic, transcriptomicmentary projects – EarlyOn and Dcode – which together form a unified platform for next-generation cancer diagnostics and biological interpretation.
EarlyOn is a liquid biopsy-based early detection program designed to address the clinical and biological heterogeneity of solid tumors. Initially, the project pioneered the combined analysis of genetic and epigenetic markers—specifically, cfDNA, DNA methylation, and miRNA—to identify robust biomarkers for early-stage cancer.
Early findings demonstrated promising diagnostic performance. However, subsequent research and emerging insights into tumor biology have underscored the need for a more comprehensive approach—one that incorporates the entire multi-omics landscape to capture the complex regulatory and signaling mechanisms underlying early tumor development.
A significant barrier in current genomic medicine remains the incomplete interpretation of functional genomics, particularly in the non-coding regions of the genome. While the majority of research has historically focused on coding regions, these comprise only a small fraction of the genome and do not account for the extensive regulatory architecture that governs gene expression and disease progression.
Understanding the Dcode is being developed as a context-aware, multi-modal machine learning model capable of modeling gene function and cancer pathway disruption across diverse omicsl context in which cancer develops. To address this, Dcode is being developed as a context-aware, multi-modal machine learning model capable of modeling gene function and cancer pathway disruption across diverse omics layers.
Unlike existing models that rely on correlation-based associations, Dcode is designed to infer functional causality by modeling how molecular disruptions propagate through genomic, epigenomic, transcriptomic, proteomic, and spatial chromatin layers. The model incorporates biological priors and spatial genome structure to enable mechanistically grounded predictions of gene function and disease relevance.
Dcode will ultimately power the biomarker discovery engine underlying EarlyOn, providing a continuous learning system that refines marker selection based on biological reasoning rather than statistical association alone. Together, EarlyOn and Dcode form an integrated platform: one focused on immediate clinical application, the other on long-term biological discovery. This dual structure combines translational impact with foundational insight into cancer biology, offering a scalable and evolving system for precision oncology.
