Single-cell transcriptomics and machine learning unveil liquid-liquid phase separation-related biomarkers in HPV-positive cervical cancer.

BACKGROUND: Liquid-liquid phase separation (LLPS) dysregulation is a key driver of oncogenesis, yet its specific role in HPV-positive cervical cancer remains undefined. We hypothesized that HPV-positive tumors exploit distinct LLPS mechanisms to evade nucleolar stress surveillance, offering a novel avenue for biomarker discovery. METHODS: We analyzed 70,519 single-cell transcriptomes across four histological conditions (Normal/Cancer, HPV+/-) to interrogate a curated panel of 120 LLPS-related genes. Our analytical framework integrated multi-tiered differential expression modeling with Random Forest classification to identify transformation-specific signatures. RESULTS: Analysis revealed a significant Cancer-HPV interaction effect characterized by the robust suppression of ribosomal proteins RPL5 and RPL11 (log₂FC ≈ - 3.4) specifically in HPV-positive cancer. Machine learning validation confirmed these genes as the top predictors (Variable Importance: RPL5 81.4%, RPL11 76.9%). The integrated model achieved 69.7% classification accuracy, predicting HPV-positive malignancy with a posterior probability of 78.4%. CONCLUSIONS: This study identifies a unique, HPV-specific suppression of the LLPS-regulators RPL5 and RPL11, mechanistically linking viral carcinogenesis to the evasion of ribosomal stress pathways. By integrating single-cell transcriptomics with machine learning, we present a novel, transformation-specific biomarker panel that distinguishes HPV-positive malignancy from both normal infection and non-viral cancer, providing new targets for diagnostic precision and therapeutic intervention.