Skip to main content
. 2020 May 28;13(9):100802. doi: 10.1016/j.tranon.2020.100802

Figure 2.

Figure 2

Differential expression analysis and feature selection identify metastasis-associated miRNA in primary and metastatic melanoma tumors. (A) Unsupervised hierarchical clustering of all primary and metastatic samples using the 90 significantly differentially expressed miRNA. Columns (samples) were clusters using Spearman’s rank correlation, and rows (miRNA) were clustered using Euclidean distance, with average linkage. Metastatic tumors are shown in pink, and primary tumors, in blue. (B) Top 20% of miRNA features selected with the highest frequency. The y-axis shows the number of feature subsets a miRNA appeared in, out of a possible 50. (C) Hierarchical clustering of all primary and metastatic samples using only 24 miRNA identified in the top 20% of selected features. All parameters the same as previously stated. (D) Confusion matrices the Ensemble RUSBoosted decision tree classifier constructed using the miRNA identified in the top 20% of feature selection. (E) ROC curve for the Ensemble model with the current classifier indicating the prediction of primary tumors.