Molecular, spatial, and AI-enabled technologies for predicting neoadjuvant therapy response in locally advanced rectal cancer.

Pretreatment prediction of neoadjuvant response remains an unresolved challenge in locally advanced rectal cancer, where management is moving towards risk-adapted and response-oriented treatment strategies. Prediction remains difficult because response reflects interacting tumour-intrinsic, immune, stromal, and spatial processes that are incompletely captured by standard clinical and radiological assessment. From a technology-centred perspective, in this Review we assess how molecular, spatial, image-based, and artificial intelligence-enabled approaches are being used to study and predict response to neoadjuvant treatment from pretreatment biopsies and how readily they may be translated into clinical workflows. Genomic profiling has its clearest current clinical role in identifying hypermutated immune-responsive subgroups, whereas broader genomic, epigenomic, and immune-repertoire markers remain largely investigational. Bulk transcriptomics has generated multiple response signatures with variable reproducibility but has consistently highlighted immune activation, interferon signalling, stromal remodelling, and epithelial-mesenchymal transition as recurrent response-associated programmes; its most feasible clinical route is likely through targeted expression panels rather than open-ended transcriptome profiling. Higher-resolution approaches such as spatial transcriptomic and proteomic platforms refine these observations by localising signals to tumour, immune, and stromal compartments and capturing therapy-induced tissue reorganisation, but they are currently best positioned for deepening biological understanding and nominating rational treatment combinations for future testing. Established platforms such as digital pathology are closer to routine deployment as they build on clinically embedded workflows, although response-prediction algorithms still require robust external validation, calibration, and regulatory-grade implementation. Patient-derived models provide functional systems for testing candidate biomarkers and resistance mechanisms. Multimodal artificial intelligence may support response prediction by identifying complementary signals across data types and distilling complex signatures into parsimonious models suitable for clinical translation. Progress will depend on harmonised, longitudinal, multi-centre studies that couple biological depth with assay standardisation, model reduction, and prospective evidence that response predictors improve patient management. © 2026 The Author(s). The Journal of Pathology published by John Wiley & Sons Ltd on behalf of The Pathological Society of Great Britain and Ireland.