Prognostic biomarker

INSIGHT: spatially resolved survival risk in stage II/III colorectal cancer

INSIGHT maps prognostic risk across H&E whole-slide images with a graph neural network, then links that risk to tumour biology through spatial transcriptomics, multiplex immunofluorescence, immunohistochemistry and bulk RNA. Trained on TCGA and SurGen and validated on the PETACC-3 phase III trial.

Read the paper (arXiv:2512.22262) · Code on GitHub · HIMICO SDK · Contact

On this page

• Overview of the INSIGHT pipeline
• INSIGHT achieves robust prognostic performance across multicentric stage II/III CRC cohorts
• Spatial risk maps recover histopathological correlates of prognosis
• Spatial transcriptomics links localised risk to epithelial identity loss, fetal-like programmes, and context-dependence
• Spatial Transcriptomic Signatures (STS)
• Signature activity per patient, with tumour genotype
• Spatial protein profiling validates epithelial, myeloid, and T-cell correlates of risk
• Patient-level transcriptomics reveals fetal-like, lineage-loss, and immune-dysfunctional programmes underlying INSIGHT risk
• Single-cell reference mapping localises INSIGHT-associated risk programmes to fetal-like epithelial and immunosuppressive myeloid states
• INSIGHT reveals molecular and immune heterogeneity within MSI-high tumours
• In MSS tumours, the risk programme replicates strongly
• A multimodal epithelial–immune risk manifold integrates spatial correlates of prognosis
• External validation of INSIGHT hypotheses
• Run INSIGHT on your own cohort with the HIMICO SDK
• Get in touch
• Read the paper, run the code