A prognostic immune signaling-related lncRNA-circRNA-mRNA network model for cervical cancer using machine learning

Cervical cancer, the third most common cancer worldwide with 50% increase in mortality, is a growing public health concern. Recent research on non-coding RNAs including long coding RNAs (lncRNAs) and circular RNAs (circRNAs) highlights their key roles in carcinogenesis across different cancers; but the studies on lncRNA-circRNA-mRNA regulatory networks (tripartite network) are still lacking in cervical cancer. This study integrates transcriptomic data to analyse circRNA-lncRNA-mRNA interactions, focusing on immune signaling pathways. To further understand the biological relevance of these interactions, differentially expressed genes were identified using DESeq2/limma, followed by functional enrichment. A competing endogenous RNA (ceRNA) network was constructed in Cytoscape, with miRNA binding serving as the central connecting factor. The immune-focused subnetwork comprised 7 mRNAs, 14 circRNAs, and 80 lncRNAs. Different machine learning prognostic models based on the ncRNA network and individual RNAs including differentially expressed coding and non-coding RNAs were developed to assess prognostic performance. Among these, the LASSO model based on the circRNA-lncRNA-mRNA network showed the best overall performance, with AUC values of 0.77 in the training set and 0.89 in the test set. Model performance was also assessed using bootstrap resampling for internal validation to ensure robustness. DLEU1, ITPR1, hsa_circRNA_101206, and miR-210 were identified as key prognostic biomarkers. Docking, HPA validation, and immune infiltration analyses confirmed the miR-210:DLEU1:ITPR1 axis as a key immune regulator in cervical cancer. This network modelling approach establishes that combinatorial modelling of coding and non-coding elements outperforms individual approaches, providing a robust framework for prognostic stratification and therapeutic targeting in cervical cancer.