Crosstalk between immune and programmed cell death and their application in colorectal cancer risk stratification.

Programmed cell death (PCD) and cellular immunity play pivotal roles in colorectal cancer (CRC) progression; understanding their crosstalk and identifying key genes could improve risk stratification and therapy. A univariate Cox analysis was performed to identify PCD-related genes (PRGs) and immune-related genes (IRGs) that are significantly associated with prognosis. Through principal component analysis (PCA), the immune and PCD levels were assessed in individual tumor tissues, subsequently categorizing patients into distinct molecular subtypes. Using the TCGA-COAD dataset as the training set and GSE17538 and GSE29621 as validation sets, machine learning model construction was performed. Leveraging the expression matrix of PRGs and IRGs, 101 machine learning models were constructed, with CoxBoost + Ridge emerging as the best-performing model. The model construction was based on previously reported methodology proposed, integrating 10 machine learning algorithms to form distinct models. The best model generated a risk score, referred to as the 'Score'. Additionally, the association between the 13 feature genes included in the Score and both prognosis and the tumor microenvironment (TME) were analyzed. Finally, experimental validation of the LTB4R gene was performed using western blotting and immunohistochemistry methods. In various datasets, PCA categorized patients into four molecular subtypes, with the immune potential index_low + PCD potential index_low group showing the highest survival rate. Among the 101 machine learning models developed, the CoxBoost + Ridge model demonstrated the best predictive performance. The Score generated by this model effectively stratified patients into high and low groups. Notably, significant differences were observed between the two groups in terms of TME, response to immunotherapy, and mutated genes. Additionally, a detailed analysis of the relationships between different feature genes, prognosis, and tumor immunity were conducted. In conclusion, the predictive model and risk score developed reflect the impact of the crosstalk between cellular immunity and PCD on the prognosis of CRC. Key genes in this crosstalk represent potential therapeutic targets, while our model provides a tool for TME evaluation and immunotherapy response prediction in CRC.