Fig. 3

Molecular stratification of HAND. a K-means clustering analysis. The three genetic features and the predicted probabilities from the six classifiers were used for k-means clustering analysis. Eight clusters were identified by inspecting the dendrogram manually. HAND-rich clusters, namely, C2, C3, C6, and C8, were renamed as H1, H2, H3, and H4, respectively. Other clusters were labeled as N. Heatmap was constructed using the ComplexHeatmap package [45]. b The representative decision tree for predicting the HAND clusters. The three genetic features were used to construct a multiclass RF classifier for the four HAND clusters and N. The number of nodes in the decision trees ranged from 13 to 29. The representative tree with the median number of nodes, 23, was shown. SVM, support vector machine; RF, random forest; GBM, gradient boosting machine; XGBL, extreme gradient boosting with linear booster; XGBT, extreme gradient boosting with tree booster; Stack, the stacked classifier