Skip to main content
. 2021 Aug 5;9:687245. doi: 10.3389/fcell.2021.687245

FIGURE 5.

FIGURE 5

The two survival subtypes of OSCC differentiated by deep learning-based model. (A) The clustering evaluation plots drawn by using the silhouette index (a) and Calinski–Harabasz criterion (b), respectively. (a) The silhouette index values for each number of clusters. The abscissa represents the number of clusters, and the ordinate represents the value of average silhouette width. The plot shows that the highest value of average silhouette width occurs at two clusters, suggesting that the optimal number of clusters is two. (b) The Calinski–Harabasz criterion values for each number of clusters. The abscissa represents the number of clusters, and the ordinate represents the value of sum of squared errors. The plot shows that when the clusters = 2, the sum of squared errors arrived at the highest values, indicating that the optimal number of clusters is two. (B) The survival differences between two subtypes, respectively, for the TCGA (a) and the two external confirmation cohorts [GSE41613 (b) and GSE42743 (b)]. The red curve represents the subtype Sub1, and the blue curve represents the subtype Sub2. The abscissa represents the follow-up time that was calculated in the number of year, and the ordinate represents the survival rate. The upper line indicates the higher survival rate, while the lower line indicates the lower survival rate. (C) The two-dimensional (2D) scatter plots showing the comparisons between deep-learning based method (a) and PCA-based method (b). The abscissa represents the dimension 1 (Dim1), while the ordinate represents the dimension 2 (Dim2). (a) The scatterplot by using top 100 principal components as supposed to 100 hidden nodes in deep learning, followed by the subsequent Cox-PH and K-means clustering. (b) The scatterplot by using deep learning based method. (D) The heatmap shows the expression patterns of ISGs–DEGs in two types of samples, i.e., samples of subtype Sub1 and Sub2. The abscissa represents OSCC samples: the rose-red samples represent 210 subtype Sub1 samples, and blue samples represent 124 subtype Sub2 samples. The ordinate represents the ISGs that were also DEGs dysregulated between the two subtypes. Pink represents the upregulated DEGs, and emerald green represents the downregulated DEGs. (E) The top list of signaling pathways shows the significantly enriched cancer-related signaling pathways that were enriched by 1,180 ISGs in the samples of subtype Sub1 and Sub2, respectively.