Proteome-anchored multi-omics propagation of recurrent somatic alterations defines tumour states in cervical cancer and enables state-guided drug repurposing

Abstract

Cervical cancer exhibits substantial molecular heterogeneity, and somatic mutation frequency alone is an incomplete surrogate for functional consequence or therapeutic vulnerability. Here, we present a proteome-anchored bioinformatics framework that integrates genomic, epigenomic, transcriptomic, and proteomic data to infer functionally expressed tumour states and to nominate candidate state-reversing perturbagens. Using matched multi-omics data from the TCGA Cervical Squamous Cell Carcinoma and Endocervical Adenocarcinoma (TCGA-CESC) cohort, recurrently mutated genes were first contextualised within the cervical cancer mutational landscape and prioritised based on regulatory DNA methylation perturbations. For selected programmes, including PIK3CA, TTN, MUC5B, SYNE1, and DST, orthogonal molecular states were independently defined at the genomic, transcriptomic, and epigenomic levels and projected onto the proteome using RPPA-based elastic-net classifiers. These models demonstrated high discriminatory performance across multiple modalities, establishing the feasibility of proteome-level encoding of upstream molecular states. State-specific, directionally signed proteomic signatures were subsequently queried against the L1000 Fireworks Display to identify perturbagens predicted to reverse inferred tumour states. Connectivity mapping revealed convergent vulnerability axes centred on stress-buffering and survival systems rather than single oncogenic drivers, including replication stress tolerance, proteostasis and secretory capacity, checkpoint control, apoptotic buffering, metabolic adaptation, and inflammatory signalling. Among the programmes examined, DST-associated states produced the most statistically robust connectivity, prioritising RTK-mTOR signalling, apoptotic priming, and metabolic-inflammatory modulation as coherent therapeutic hypotheses, while other programmes yielded more moderate signals consistent with hypothesis generation. Collectively, this study establishes a scalable, proteome-anchored multi-omics framework that moves beyond mutation-centric analysis toward functional tumour-state inference and provides a principled route for state-guided therapeutic hypothesis generation in cervical cancer.