Constructing the immune-related signature of colon cancer

We exerted Wilcoxon signed-ranked tests to screen differentially expressed immune-related genes between normal tissue samples and primary tumor tissue samples from TCGA, with — FC (Fold change) —>1 and false discovery rate (FDR) < 0.05. The patients with colon cancer from TCGA were randomly divided into a training set (N = 224) and a testing set (N = 221), using a R package called “caret”. For training set, univariate Cox proportional hazard model (CoxPH) was used to screen immune-related genes with prognostic value (P < 0.05). The least absolute shrinkage and selection operator (LASSO) regression model (iteration = 1,000) with an elastic-net penalty was performed for further screening, using a R package called “glmnet” (Friedman, Hastie & Tibshirani, 2010). Then multivariate CoxPH was performed to screen out genes which were used to construct the immune-related signature. The signature gave patients in both training and testing set risk scores based on model coefficients of multivariate CoxPH:

Patients were classified into low- and high-risk group based on the median of risk scores.