Treatment classification with Random Forests.

A. Principal Component Analysis (PCA) plots from all profiles based on a set of 1338 genes that were differentially expressed in at least one condition (disease or treatment) against wild-type (WT) control samples. B. PCA plot on the same profiles based on the top 100 most important genes based on Random Forest Classification. C. Confusion matrix of sample classification according to the best Random Forest Model. Rows correspond to the actual condition of origin. Columns correspond to the condition to which the sample is assigned by the model. Numbers inside cells represent number of samples assigned to each pair. D. Top 30 most important genes according to the best RF model ranked on the basis of mean decrease in Gini coefficient. RF model was trained on all samples.