FIGURE 4.

Construction of a classifier to predict immune subgroups using machine-learning algorithms. (A) Machine-learning algorithms were used to develop a simple predictor to predict immune clusters, thereby randomly assigning all samples to the group with poor or good prognosis until the best prediction efficiency is obtained. A total of 26 hub immune genes were identified for the prognostic predictor for subgroup classification. (B) For the training set samples, immunophenotyping clustering was used to determine whether the samples belonged to Cluster A or Clusters B and C. (C). Survival difference between immunophenotyping clusters of ccRCC patients in the training cohort. (D) The logistic regression coefficient was further used to calculate the risk score of each sample.