Fig. 3

Screening of hub gene via multiple machine-learning algorithms. a–c Identification of hub gene by RF. Distribution of out-of-band (OOB) error rate at various values of mtry (a) and trees (b). Variable importance, as measured by the mean decrease in accuracy (left panel) or the Gini coefficient (right panel), is computed using the OOB error (c). Genes are shown in descending order of importance. d, e Establishment of hub gene by LASSO logistic regression analysis. LASSO coefficient profile of the 8 genes (d). Selection of the optimal parameter (lambda) in the LASSO model, and generation of a coefficient profile plot (e). f–j Process of WGCNA. Analysis of network topology for various soft-thresholding powers (f, g). The x-axis reflects the soft-thresholding power. The y-axis reflects the scale-free topology model fit index (f) and the mean connectivity (g). Clustering dendrogram of differentially expressed genes related to TSCC, with dissimilarity based on topological overlap, together with assigned module colors (h). Module-trait associations (i). The gene significance for TSCC in the turquoise module (j). k Venn diagram shows the intersection of the hub gene obtained by the three strategies