A novel machine learning-based immune prognostic signature for improving clinical outcomes and guiding therapy in colorectal cancer: an integrated bioinformatics and experimental study.

Immune cells are pivotal components in the tumor microenvironment (TME), which can interact with tumor cells and significantly influence cancer progression and therapeutic outcomes. Therefore, classifying cancer patients based on the status of immune cells within the TME is increasingly recognized as an effective approach to identify prognostic biomarkers, paving the way for more effective and personalized cancer treatments. Considering the high incidence and mortality of colorectal cancer (CRC), in this study, an integrated machine learning survival framework incorporating 93 different algorithmic combinations was utilized to determine the optimal strategy for developing an immune-related prognostic signature (IRPS) based on the average C-index across the four CRC cohorts. Notably, IRPS was demonstrated to be an independent risk factor for predicting the survival outcomes of CRC patients, showing superior performance compared to traditional clinical features and 63 published signatures in both training and validation cohorts. Furthermore, CRC patients classified in the low-risk group according to the IRPS showed higher sensitivity to immunotherapy than those in the high-risk group, suggesting that low-risk patients are more likely to benefit from immunotherapy. Through in silico screening of potential compounds, dasatinib, vinblastine, and YM-155 were identified as potential therapeutic agents for high-risk CRC patients. In vitro studies demonstrated that knockdown of APCDD1, a key component of the IRPS, inhibited the proliferation, migration and invasion of HT-29 cells and promoted their apoptosis. Thus, the IRPS serve as a powerful tool for predicting patient prognosis, immunotherapy response and candidate drugs, thereby enhancing clinical decision-making and treatment evaluation of CRC.