Fig. 8

Stability assessment of the unsupervised clustering of the adjusted RAS matrices. UMAP dimensionality reduction with HDBSCAN clustering was performed on 21 samples of the GN/NB dataset over 1000 iterations. UMAP’s parameter n_neighbor attribute was set to 3. The subplots show 50 random selected iterations of UMAP followed by HDBSCAN clustering (x-axis) for A) \(M_{W}\) (RAS values), B) \(M_{W_{1}}\) (RAS values adjusted by TP), C) \(M_{W_{2}}\) (RAS adjusted by the path-specific TP), and D) \(M_{W_{3}}\) (RAS adjusted by the recursive TP). The color indicates the cluster of the respective sample. GN is represented through a circle, while NBs are rectangular. The same samples are connected with lines between the iterations, indicating cluster stability. The y-axis represents the sum of x- and y-coordinates of the UMAP per sample