A Clinical Evidence-Guided Bioinformatics Framework for Expanding Cisplatin-Based Combination Therapies in Cervical Cancer

Background Cisplatin-based combination therapy remains the foundation of systemic treatment for advanced cervical cancer; however, identifying effective therapeutic partners is challenging because of tumour heterogeneity and the large number of possible drug combinations. Existing computational platforms primarily evaluate experimentally generated drug combinations and provide limited support for systematically expanding clinically established treatment regimens while integrating mechanistic evidence to explain candidate prioritisation. Methods We developed a clinical evidence-guided computational decision-support and hypothesis-generation platform for systematic expansion of cisplatin-containing cervical cancer therapies. Clinically investigated cisplatin combination regimens were identified from ClinicalTrials.gov and used as pharmacological reference backbones. Model-specific dose–response profiles were reconstructed from Genomics of Drug Sensitivity in Cancer (GDSC2) data and evaluated using Bliss independence, Highest Single Agent, Loewe additivity, and Zero Interaction Potency interaction models. Candidate third-drug additions were systematically prioritised relative to the pharmacological performance of each reference regimen. Pharmacological findings were subsequently integrated with Cancer Dependency Map CRISPR gene dependency, transcriptomic expression, and somatic mutation data, followed by graph-learning analyses and patient-level molecular contextualisation using the independent GSE299125 cervical cancer transcriptomic cohort. Results The framework demonstrated substantial model-specific heterogeneity in the pharmacological performance of clinically established cisplatin regimens and identified distinct candidate third-drug partners across cervical cancer models. Integrating functional genomic dependency, transcriptomic expression, and genomic context provided mechanistic interpretation beyond pharmacological interaction scores alone, while graph-learning analyses captured systems-level relationships among prioritised therapeutic candidates. Patient-level projection further demonstrated that prioritised molecular targets were represented across independent cervical cancer transcriptomic profiles, providing biological contextualisation of experimentally derived therapeutic hypotheses. Conclusions This study presents a clinically grounded, explainable computational framework that extends conventional drug-combination analysis from retrospective synergy assessment toward evidence-guided therapeutic optimisation. By integrating clinical trial evidence, reconstructed pharmacology, functional genomics, graph learning, and patient-level molecular contextualisation within a unified workflow, the platform provides a reproducible foundation for therapeutic hypothesis generation, systems pharmacology, and precision oncology research in cervical cancer. Implementation and Availability: The platform is implemented as an interactive R Shiny application. A publicly accessible prototype is available through Posit Connect Cloud, and the complete source code and supporting datasets are available through the accompanying GitHub repository.