Network-Based Stratification Refines Stratification of Intermediate-Risk Acute Myeloid Leukemia Samples

The European LeukemiaNet (ELN) risk stratification of acute myeloid leukemia (AML) uses genetic and molecular markers to categorize patients. However, disease heterogeneity, particularly in the intermediate-risk group, complicates stratification. Ideker et al. developed the Network-Based Stratification (NBS) method, combining protein network analysis and mutation profiling via machine learning. We applied NBS to intermediate- risk AML patients to refine prognosis and identify distinct molecular subtypes compared to the 2022 ELN scheme. We selected 170 intermediate-risk AML patients based on the 2022 ELN classification from TCGA (n=58), BEAT AML (n=87), and FIMM (n=25) datasets. Using NBS, we analyzed 3,108 genes from WGS or WES data, mapping them onto a cancer-specific protein network for clustering based on network-propagated mutation profiles. We conducted 200 iterations of sub-sampling, considering patients with at least 3 mutated genes and using consensus clustering for robust stratification, assessing associations with clinical and transcriptomic features. NBS identified five distinct molecular subgroups characterized by unique mutation patterns: IDH1-dominant (Cluster 1), DNMT3A-dominant (Cluster 2), low-frequency multi-mutated (Cluster 3), FLT3/NPM1/DNMT3A co- mutated (Cluster 4), and FLT3-dominant (Cluster 5). Cluster 4 showed significantly worse overall survival (HR = 1.81; p = 0.05). In addition, ex vivo drug sensitivity and transcriptomic analyses revealed significant variation in therapeutic response and pathway activation across clusters. These findings underscore the power of machine learning–driven approaches like NBS to uncover hidden molecular structure within intermediate-risk AML groups, enabling more precise prognostication and potentially informing personalized therapeutic strategies.