Fig. 5

Exploration of the signature genes by random forest classifier and neural network model. (A) The relationship plot between the error rate and the number of decision trees. When the number of decision trees is about 400, the error rate is relatively stable. (B) The importance score of the signature genes using the Gini coefficient method. The X-axis denotes the signature genes, and the Y-axis denotes the importance index. (C) The neural network diagram demonstrates the relationship between the signature genes and the two pyroclusters. (D) The signature genes were differentially expressed in the two pyroclusters. (E-F) The receiver operating characteristic curve analysis of the classification efficiency of the signature genes in training sets (TCGA-HNSCC and GSE65858 cohorts) and external testing sets (GSE3292 and GSE6792 cohorts). (G) Expression levels of the 6 signature genes in HPV-positive HNSCC and their adjacent normal tissues using qRT-PCR.