Fig. 2From: Using protein-per-mRNA differences among human tissues in codon optimizationRandom Forest models identify two clusters of human tissues with distinct codon signatures. A Receiver operating characteristic (ROC) curves of lung and kidney random forest classifiers, in which the codon usage of genes is used to predict whether they are high-PTR or low-PTR in the respective tissue (see the “Methods” section). B Ratios of the codon usage between high-PTR and low-PTR genes in each tissue. Codons and tissues are hierarchically clustered using Euclidean distances and the complete-linkage method. The barplot on the left shows the mean AUC of the ROC curve of the RF model of each tissueBack to article page