Construction of a deep learning model and identification of BSG, PPARD, and SLC16A8 expression as potential indicators in the context of strategies for precision therapy to acute myeloid leukemia.

OBJECTIVE: Acute myeloid leukemia (AML) exhibits significant heterogeneity and aggressiveness. This study aimed to investigate T cell heterogeneity in the AML tumor microenvironment using single-cell RNA sequencing (scRNA-seq) and identify potential biomarkers for prognosis and precision therapy.

METHODS: scRNA-seq data from AML patient samples were analyzed to identify T cell subsets. A prognostic risk model was constructed using random forest and LASSO regression analyses based on key genes derived from a specific T cell cluster (Cluster 4). The model's predictive performance was validated using external datasets.

RESULTS: Analysis revealed significant functional heterogeneity among T cell subsets. Cluster 4 T cells showed distinct gene set activities related to immune regulation. Three genes - BSG, PPARD, and SLC16A8 - were identified as independent prognostic factors. The risk model effectively stratified patients into high-risk and low-risk groups, with the high-risk group demonstrating significantly poorer survival outcomes. The model showed robust predictive accuracy, with areas under the ROC curve of 0.78, 0.86, and 0.86 for 1-, 3-, and 5-year survival, respectively.

CONCLUSION: This study highlights the functional diversity of T cells in AML and identifies BSG, PPARD, and SLC16A8 as promising biomarkers for prognostic stratification. The developed risk model provides a valuable tool for guiding personalized treatment strategies in AML.