Development and validation of a five-gene palmitoylation-associated prognostic model in acute myeloid leukemia

Abstract

To develop and validate a prognostic model based on palmitoylation‑related genes for risk stratification in acute myeloid leukemia (AML), and to explore its associations with the immune microenvironment, drug sensitivity, and single‑cell expression patterns. A five‑gene prognostic signature was constructed using univariate Cox regression and LASSO‑Cox analyses in the TCGA‑LAML training cohort (n = 131) and validated in an independent GEO cohort (GSE71014, n = 104). Risk scores were calculated, and patients were divided into high‑ and low‑risk groups. Overall survival (OS) was compared by Kaplan‑Meier curves, and model performance was evaluated by time‑dependent ROC. Immune cell infiltration, immune checkpoint expression, and drug sensitivity (predicted IC50) were analyzed. Single‑cell RNA‑seq data (GSE116256) were used to examine cell‑type specific expression of the model genes. The five‑gene model (LYPLA2, ZDHHC11, ZDHHC7, ABHD6, ABHD17C) significantly distinguished high‑risk patients with poorer OS in both training (P < 0.001) and validation (P < 0.001) sets, with AUCs of 0.72, 0.73, 0.75 (training) and 0.67, 0.70, 0.60 (validation) for 1‑, 3‑, and 5‑year survival. Multivariate Cox regression confirmed independent prognostic value (HR = 2.78, P < 0.001). High‑risk patients showed increased regulatory T cells, activated dendritic cells, and follicular helper T cells, along with positive correlations with PD‑1, TIM‑3, LAG3, and VISTA expression. Ninety‑eight compounds with differential predicted IC50 values were identified. Single‑cell analysis revealed LYPLA2 and ZDHHC7 enrichment in monocytes/macrophages and dendritic cells. The five‑gene palmitoylation‑associated prognostic model effectively stratifies AML patients by risk, reflects immune microenvironment dysregulation, and offers potential guidance for personalized therapy.