Beyond prediction: AI as a mechanistic microscope and digital twin for colorectal cancer immunotherapy

Colorectal cancer (CRC) remains a major cause of cancer-related death, yet the benefits of immune checkpoint inhibitors are limited to a small subset of patients, particularly those with microsatellite instability-high or mismatch repair-deficient tumors. Most patients with microsatellite-stable disease derive little benefit, and even responsive subgroups show substantial heterogeneity and acquired resistance. These challenges highlight the need for biomarkers and therapeutic frameworks that can not only predict response, but also explain underlying biology and support dynamic treatment decisions. In this review, we propose that artificial intelligence (AI) can move beyond prediction to serve two broader roles in CRC immunotherapy: as a mechanistic microscope that reveals hidden tumor-immune interactions from multimodal data, and as a digital twin that models patient-specific therapeutic trajectories over time. We summarize recent advances in AI-based pathology, imaging, and liquid biopsy for pretreatment stratification and response monitoring, and discuss how these approaches may inform resistance mapping, adaptive trial design, and strategies to convert immunologically "cold" tumors into "hot" tumors. We further examine key translational barriers, including generalizability, interpretability, and regulatory validation. By integrating multimodal data with mechanistic modeling, AI may help shift CRC immunotherapy from population-level prediction toward dynamic, individualized precision oncology.