Fig. 1

Transcriptome-based WGCNA analysis identified hub lncRNA CARMN related to the NOR-CIN-CC transition. a Heatmaps of expression levels of mRNAs (left) and lncRNAs (right) by RNA-seq in 10 CC, 10 adjacent normal tissues (NOR), and 10 CINs. b A total of 34 modules obtained by WGCNA analysis with a correcting cutHeight = 0.25 by the Dynamic Tree-Cut algorithm. c Associations of modules with groups and ages. d-e KEGG pathway analysis of hub mRNAs in the Blue (d) and Lightyellow (e) Module. f CARMN-TF-mRNA network in the two modules by WGCNA. Blue color indicates the molecule belongs to the blue module, while yellow color belongs to the lightyellow module. The circles represent mRNAs, the squares represent lncRNAs, and the red oblique square represent transcription factors. g The RNA-seq group comparison showed that red dots represented upregulated genes, blue dots represented downregulated genes, and black dots indicated no significant difference between the two groups. h Nine key lncRNAs involved in the cervical carcinogenesis process were identified by intersecting differentially expressed lncRNAs with those in key modules from WGCNA. i The PCA results showed that the sample characteristics were highly consistent across datasets after batch correction. j The ROC curves for each group were presented using ten different machine learning algorithms. k-m Lightgbm (k), KNN (l), and XGBOOST (m) demonstrated the best performance in predicting sample groups, with variable importance explained based on SHAP values. (n) Validation of CARMN expression in the original 30 and additional 83 cervical tissues. From left to right: RNA-seq data, qPCR results of original 30 cervical tissues, cervical adenocarcinomas (CAC, 20 tumors vs. 20 NOR), and additional tissues of cervical squamous cell carcinomas (CESC, 23 tumors vs. 20 NOR). o RNAScope results showed the abundance of CARMN in tumor and normal tissues. ns, P ≥ 0.05; *, P < 0.05; **, P < 0.01; ***, P < 0.001