Reconstructing cervical cancer tumour microenvironment states from DNA methylation profiles using a machine learning framework

Background: Cervical cancer burden is high in LMICs, worsened by HPV and HIV co-infection, yet frameworks linking bulk epigenomic data to tumour microenvironment (TME) states remain limited. Methods: A DNA methylation-based ML framework integrates feature selection, deconvolution, unsupervised clustering, and predictive modelling to reconstruct TME states. Results: Four distinct TME states were identified, differing in immune activity, stroma, epithelial differentiation, and proliferation. HIV-associated perturbations were largely restricted to immune-inflamed tumours. Inferred states enabled alignment of patient tumours with cell-line models. Conclusions: The framework supports biomarker discovery, risk stratification, and bridging patient biology with preclinical models.