Integrated machine learning-based establishment of a prognostic model in multicenter cohorts for acute myeloid leukemia.

BACKGROUND: Acute myeloid leukemia (AML) is a highly heterogeneous malignancy, with leukemic cell diversity contributing to disease progression and treatment resistance. This study aimed to evaluate the functional and prognostic significance of leukemic cell-related genes. METHODS: We analyzed single-cell RNA sequencing data to identify malignant marker genes in AML. Consensus clustering was used to assess associations with prognosis and immune responses. A prognostic model, the malignant leukemia marker gene prognostic signature (MLAPS), was developed using 101 models across 10 machine learning algorithms and validated in five independent cohorts. Functional assays were conducted to explore the role of CD69. RESULTS: We identified a set of malignant marker genes significantly correlated with prognosis and immune classification. The MLAPS showed strong predictive performance, surpassing most clinical features and previously published signatures. Experimental validation confirmed that CD69 promotes malignant progression in AML. CONCLUSION: This study highlights the clinical value of leukemic cell-specific genes and presents MLAPS as a robust prognostic tool. CD69 may serve as a potential therapeutic target in AML.