摘要：SummaryPredicting colorectal cancer recurrence after tumor resection is crucial because it promotes the administration of proper subsequent treatment or management to improve the clinical outcomes of patients. Several clinical or molecular factors, including tumor stage, metastasis, and microsatellite instability status, have been used to assess the risk of recurrence, although their predictive ability is limited. Here, we predicted colorectal cancer recurrence based on cellular deconvolution of bulk tumors into two distinct immune cell states: cancer-associated (tumor-infiltrating immune cell-like) and noncancer-associated (peripheral blood mononuclear cell-like). Prediction model performed significantly better when immune cells were deconvoluted into two states rather than a single state, suggesting that the difference in cancer recurrence was better explained by distinct states of immune cells. It indicates the importance of distinguishing immune cell states using cellular deconvolution to improve the prediction of colorectal cancer recurrence.Graphical abstractDisplay OmittedHighlights•Distinct immune cell states predict colorectal cancer recurrence•Methylation patterns of immune cells altered after tumor infiltration•Combining immune cell states and clinical factors improves recurrence prediction•The proportion of TIIC-like DCs is a crucial factor for the recurrence predictionHealth sciences; Health informatics; Oncology; Immunology; Bioinformatics; Biocomputational method; Systems biology; Cancer systems biology
