Figure 3.

Top 30 gene signatures and clinicopathological relevance selected by MRMR approach. (A) Unsupervised hierarchical clustering of diagnostic top-ranked 30 gene signatures effectively discriminated between the ESCC group (n = 71) and the control group (n = 80). The 30-gene signature enables non-random clustering of all samples (p < 0.001, Fisher’s exact test). (B) Obviously different distributions of principal components analysis (PCA) between the control and ESCC group. Values of principal components were calculated using prcomp package in Rstudio. (C) Obviously different distributions of t-SNE visualization between the ESCC and control group. Values of t-SNE signature were calculated using Rtsne package in Rstudio. (D) Positive and (E) negative correlations between the diagnostic genes and clinical data. p-value above 0.05 was identified significantly correlated based on Pearson method between continuous data and Spearman method if one of the variables was not. Plots were drawn using the RCircos package in Rstudio. ESCC, esophageal squamous cell carcinoma; PCA, principal components analysis; t-SNE, t-distributed stochastic neighbor embedding.