Integrated single-cell and machine learning analysis identifies an autophagy- and immune-related prognostic signature in adrenocortical carcinoma

Adrenocortical carcinoma (ACC) is an uncommon but highly aggressive endocrine malignancy characterized by poor survival outcomes and remarkable molecular heterogeneity. Emerging evidence suggests that autophagy participates in tumor progression and immune regulation; however, its comprehensive molecular landscape and clinical implications in ACC remain largely undefined. Multi-omics datasets from TCGA-ACC, E-GEOD-19,776, and GSE252112 were integrated to identify core autophagy- and immunity-associated genes. A prognostic signature was established through machine-learning-based model selection. Survival analysis, ROC curves, and nomogram construction were performed to evaluate predictive performance. Immune characteristics were explored using ssGSEA, CIBERSORT, TIP, ESTIMATE, and TIDE algorithms. Somatic mutation profiling, pathway enrichment analysis, and oncoPredict-based drug sensitivity assessment were further conducted. A three-gene signature composed of BIRC5, CASP3, and MAPK8 was developed and showed generally consistent prognostic stratification across the available cohorts. Individuals classified as high risk exhibited significantly shorter overall survival, increased tumor mutation burden, frequent TP53 alterations, and distinct immune-context features inferred from bulk transcriptomic deconvolution analyses. Functional analyses indicated enrichment of cell-cycle regulation, epithelial–mesenchymal transition, and DNA replication pathways. Moreover, predicted IC50 values indicated potential differential sensitivity to several targeted agents in the high-risk subgroup. This autophagy- and immune-related signature may provide a useful tool for prognostic assessment in ACC and may reflect potential associations with genomic instability, immune-context variation, and therapeutic vulnerability, which require further experimental and clinical validation.