HELIOS: A spatial omics-informed multi-organ cell atlas enables AI-based tumour microenvironment profiling and biomarker discovery from routine histopathology

Spatial omics defines tumour microenvironments with high biological resolution but remains difficult to scale across large clinical cohorts. We developed HELIOS (H&E Learning Informed by Omics and Spatial Profiling), which transfers spatial-omics-derived cell identities into routine haematoxylin and eosin histology for interpretable single-cell tumour microenvironment profiling. HELIOS integrates 12.9 million cells from 60 specimens across 13 tissue types and four spatial profiling platforms, spanning eight epithelial, immune, stromal, and vascular populations. The resulting multiclass detector generalised across independent histopathology benchmarks. Applied to colorectal, breast, and lung cancers, HELIOS recapitulated molecular subtype-associated microenvironment states and identified prognostic immune and stromal phenotypes. Tumour-infiltrating lymphocytes were consistently associated with favourable outcomes, whereas neutrophil and fibroblast phenotypes showed cancer-specific associations. HELIOS-derived features remained prognostic after adjustment for clinicopathological and molecular variables. These findings demonstrate that spatial omics can provide scalable biological supervision for extracting clinically relevant tumour microenvironment features from routine H&E, supporting scalable risk stratification and biomarker development across clinical cohorts and retrospective pathology archives.