Predictive features for relapse in adult AML detected by machine learning analysis. (A) Relationships between copredictive features associated with diagnosis (left) and relapse (right) among adult AML cases are visualized by using VisuNet. The color of the nodes shows the expression level, with 3 bins for high (orange), medium (gray), and low (blue) expression. The rule support is indicated by the size of the respective node, whereas the support for each connection is visualized by the thickness and color of the connective line. Rules were filtered according to a false discovery rate <0.05. (B) Scatter plots with mean and standard deviation illustrating the log2-transformed, TMM-normalized expression values in diagnosis and unpaired relapse samples for CD6, INSR, and ZNF773. The borders of the 3 bins, corresponding to low, medium, and high expression of the respective genes, are indicated by green grid lines on the y-axis. Applied statistical test, Mann-Whitney test. Supplemental Table 10C provides details regarding samples included in this figure.