Figure 3
From: NBIA: a network-based integrative analysis framework – applied to pathway analysis
Kaplan-Meier survival analysis of AML subtypes discovered by consensus clustering (A panels), hierarchical clustering (B panels), local shrinkage (C panels), and cluster ensemble (D panels). The top panels show AML subtypes discovered using genome-wide expression values (all genes) while the bottom panels shows the subtypes discovered using genes selected by NBIA. In each panel, the colored curves represent the survival probability of different subtypes. For any of the four methods, we are not able to find subtypes with significantly different profiles when using the genome-wide expression values. In contrast, when applied in conjunction with the pathway signature obtained from NBIA, any of the four methods identifies subtypes with very significant survival profiles. Interestingly, there is one group of patients that are always grouped together in the high-survival group using CC, HC, and local shrinkage. When performing an ensemble of the three partitioning, we are able to separate this group from the rest (panel D). The cluster ensemble algorithm identifies two groups of patients that have very different survival profiles (Cox p-value 2 × 10−6). Among the high-survival group, almost 60% of the patients survived at the end of the study (more than 8 years). In contrast, only approximately 10% of the other group survived at the end.
