A knowledge graph-guided transcriptomic framework identifies a 20-gene prognostic signature for cervical cancer

Abstract Background Current cervical cancer risk stratification relies on FIGO stage and lymph node status, tools that miss molecular heterogeneity and leave patients over- or under-treated. Conventional transcriptomic biomarker discovery uses differential expression alone, returning thousands of candidates—most of which are pan-cancer associations that lack disease specificity and fail independent validation. We developed a two-filter framework that combines knowledge graph (KG) embedding specificity with transcriptomic differential expression to identify cervical cancer-specific prognostic genes. Results Among ten embedding models for prediction of gene-disease association, ComplEx showed the highest cervical cancer specificity (Top 100 mean Z-score = 4.19, range: 3.54–6.71). Intersection of KG-predicted candidates (Z ≥ 1.645,P < 0.05) with DESeq2 differentially expressed genes yielded 476 high-confidence candidates. From TCGA-CESC (n = 250, 50 deaths), a 22-gene signature was derived by cross-validated Top-K selection (CV C-index = 0.776). The signature genes classified patients into high- and low-risk groups (log-rank P = 6.33 × 10⁻¹⁰, bootstrap mean C-index = 0.816, 95% CI: 0.762–0.865). In multivariable cox analysis adjusting for age and FIGO stage, each standard-deviation increase in risk score was associated with a 123% increase in mortality risk (HR = 2.23, 95% CI: 1.67–2.96, P < 0.0001). The signature maintained discrimination in Stage I patients (C-index = 0.815) and showed sustained temporal accuracy (mean 5-year AUC = 0.829). Permutation testing confirmed non-random discrimination (P < 0.0001). In GSE52903, all 22 genes mapped successfully and the signature showed significant differential expression between tumor and normal tissue (P = 0.042). Conclusions This study introduces the Z-Score-Based Specificity Framework to integrative KG specificity scoring with transcriptomics analysis for cervical cancer biomarker discovery. The 22-gene signature demonstrates strong internal validity, sustained temporal discrimination, and independent biological coherence. Its high performance in early-stage disease suggests potential utility for guiding adjuvant therapy decisions and show the potential of KG-powerd