Bias-Controlled Computational Prioritisation of Therapeutic Hypotheses From Heterogeneous Molecular Data Identifies PLK1 as the Leading MTX-Combination Candidate in Choriocarcinoma

Background: Methotrexate (MTX) resistance remains clinically relevant in gestational trophoblastic neoplasia and choriocarcinoma. Because disease-specific molecular cohorts are limited, computational prioritisation of therapeutic hypotheses from heterogeneous public data must explicitly distinguish experimental prioritisation from therapeutic validation. Methods: Protein and phosphoprotein changes from an MTX-resistant JEG-3 model were used as effect-size seeds for random walk with restart propagation across four molecular interaction networks. Computational robustness was assessed using dropout analyses, degree-preserving edge randomisation, and seed-degree-matched permutation. Orthogonal evidence from DepMap/PRISM, external transcriptomic datasets, DrugComb records, a normal-placenta single-cell reference, and normal-tissue RNA expression was integrated to evaluate biological plausibility and translational constraints. Candidate hypotheses were prioritised using transparent multi-criteria decision analysis (MCDA), and cross-modal agreement was summarised descriptively using rank concordance rather than a combined-probability test, because the evidence sources share overlapping cell-cycle and DNA-damage-repair signal and therefore do not satisfy the independence assumption required by such tests. Results: PLK1 ranked first across all 3000 MCDA perturbations, demonstrating stable prioritisation despite uncertainty in evidence weighting. Network propagation supported prioritisation but not discovery-level significance: the edge-degree-preserving empirical p value was 0.017 (BH-FDR, 0.479), and the seed-degree-matched p value was 0.048 (BH-FDR, 0.896). No candidate reached FDR significance in external transcriptomic meta-analysis or lineage- and proliferation-adjusted DepMap/PRISM regression. PLK1 was the only candidate ranked first across all four evaluated evidence analyses (both network null models, adjusted DepMap analysis, and external transcriptomic meta-analysis), indicating complete descriptive cross-modal rank concordance without assuming statistical independence across analyses. PLK1 also showed strong but nonselective basal dependency across three choriocarcinoma models. DrugComb contained one independent MTX plus PLK1-inhibitor observation with positive ZIP synergy, and the exploratory tumour-to-highest-critical-normal RNA ratio was 1.25. Conclusions: This bias-controlled computational prioritisation framework identifies MTX plus PLK1 inhibition as the highest-priority therapeutic hypothesis for experimental evaluation in choriocarcinoma. PARP1 and PRKDC remain secondary biomarker-stratified hypotheses. More broadly, the framework demonstrates how heterogeneous molecular data can be integrated into transparent, reproducible therapeutic hypothesis prioritisation while explicitly separating computational prioritisation from therapeutic validation.