Figure 2

From discrete classification to grades of membership. a. The same correlation matrix as in Figure 1b but with samples re-grouped so that the Type 1 samples are on the lower left side, showing that samples can be classified into two main types of expression patterns. b. The same samples can be subdivided into three classes, and re-ordered accordingly in the correlation heatmap, still showing reasonably strong distinction between the three classes. c. The same heatmap re-ordered by the first Principal Component scores, resulting in a gradual progression from one prototype to the other.