Graph-based method reveals key genes and their distinct roles across acute myeloid leukemia subtypes

Abstract Objectives Acute myeloid leukemia (AML) is a heterogenous hematologic malignancy affecting both children and adults, characterized by diverse molecular alterations and variable clinical outcomes. Identifying subtype-specific gene markers is therefore important for improving disease characterization and guiding personalized therapeutic strategies. In this study, we aimed to identify potential genes involved in AML using a graph-based clustering method and to evaluate their differential expression across AML subtypes. Methods A protein-protein interaction network (PPIN) was constructed and analyzed using the DPClusOST clustering algorithm to identify protein clusters at densities ranging from 0.1 to 1.0. Cluster significance was evaluated using a significance score (SScore) based on Fisher’s exact test, and clusters containing AML gene panels were assessed through receiver operating characteristic (ROC) analysis. The cluster density with the highest area under the curve (AUC) was subsequently subjected to differential gene expression and pathway enrichment analyses. Results At a clustering density of 1.0, 32 clusters were identified, including 16 novel AML-associated genes that exhibited significant differential expression across AML subtypes (e.g., PML-RARα and CD34; |log 2 FC| ≥1, adjusted p<0.05). Pathway enrichment analysis further demonstrated significant involvement in AML-related biological processes (p-value <0.05), including PI3K/AKT signaling and hematopoietic dysregulation, such as refractory macrocytosis, abnormalities of bone marrow stromal cells, and multiple lineage myelodysplasia, inferring the potential involvement of the newly discovered genes in the clonal, acquired stem cell disorder characterization of AML. Conclusions These findings demonstrate that graph-based clustering can effectively identify novel AML-associated genes with potential relevance for subtype characterization.