Integrating machine learning and multi-omics analysis to explore Treg-associated programmed cell death features in...

Treg infiltration and programmed cell death are important factors influencing cancer progression, and they interact with each other. However, the significance of Treg-related programmed cell death (PCD) characteristics in clear cell renal cell carcinoma remains unclear.

Through Mendelian randomization, we identified PCD genes and Treg markers that are highly associated with ccRCC outcomes. Subsequently, based on Treg-related PCD genes, we constructed a diagnostic model utilizing a multi-layer perceptron (MLP) and integrated 10 machine learning algorithms to construct a prognostic model, which was then explained by the SHAP method. After exploring functional differences and chemotherapy sensitivity differences between high- and low-risk groups in the prognostic model, we validated the core gene of the model through in vitro cell experiments. Finally, we screened molecular drugs targeting the core genes using the DSigDB database and performed molecular docking and molecular dynamics validation.

Utilizing Mendelian randomization (MR), we first established causal links between specific Treg subtypes and PCD gene CASP9 with renal cancer outcomes. Leveraging shared Treg-PCD molecular features, we developed a MLP-based diagnostic model achieving an AUC of 0.987 in external validation. Further, a robust prognostic index Treg-Programmed Cell Death Score (TPCDS) was constructed using 101 machine learning combinations, demonstrating superior stratification across multi-cohort data. High TPCDS correlated with immunosuppressive microenvironments including increased Tregs, T-cell exhaustion, HLA downregulation and poor immunotherapy response, while guiding chemotherapy sensitivity. Functional assays confirmed the core gene SLC11A1 as an oncogenic driver promoting proliferation, migration, and invasion. Molecular docking and dynamics simulations identified Atovaquone as a high-affinity inhibitor of SLC11A1.

We explored the significance of Treg and programmed cell death characteristics in the ccRCC tumor microenvironment and established clinically translatable tools for ccRCC diagnosis, prognosis, and personalized therapy selection, thus promoted the application of explainable machine learning models in precision oncology. Furthermore, We have identified SLC11A1 as a highly promising therapeutic target for ccRCC.