Glioblastoma remains one of the most lethal primary brain tumors, with standard-of-care treatment offering limited survival benefit. Current targeted therapies have repeatedly failed clinically because tumors exploit alternative signaling pathways to circumvent inhibition. Therapeutic vulnerabilities in GBM are more robustly captured at the pathway level than at individual gene mutations, motivating a systems-level approach to drug prioritization.
We present a Pathway-Drug Coexpression Network (PDxN) pipeline connecting biological pathway gene sets to drug-induced transcriptomic signatures from large-scale perturbational genomics data. Compounds whose transcriptomic effects most strongly invert a tumor's dysregulated pathway state are prioritized, shifting focus from mutational profiling to functional reversal of disease-associated transcriptomic states. Pathway signatures are derived from molecularly stratified, patient-derived ex vivo GBM cultures representing four driver subtypes: EGFR, FGFR, MDM2, and MEK, with differential pathway activity analysis capturing the transcriptomic basis of sensitivity and resistance per subtype.
A key methodological advance is correction of a systematic bias in which compounds with disproportionately large numbers of database entries dominated prioritization outputs independent of biological relevance. Upstream quality filtering and downstream score aggregation eliminate this artifact, grounding predictions in reproducible transcriptional signal rather than sampling imbalance.
Validation against dose-response data from 500 drugs tested on patient-derived GBM cultures demonstrates improved predictive specificity following bias correction. Predictions are enriched for empirically active compounds at the top of ranked lists, with Precision@K10 of 0.60–0.70 for EGFR and approaching 0.80 for MEK, substantially exceeding permutation-based baselines. AUC values confirm consistent discriminative performance across benchmarks, and mechanism of action enrichment identifies convergent therapeutic themes including EGFR, CDK, and MTOR inhibitors across molecular subgroups.
This work delivers a validated drug prioritization framework and ranked repurposed candidates for treatment-resistant glioblastoma, within a modular architecture applicable to other rare and heterogeneous cancers where pathway-level transcriptomic analysis offers a tractable route to drug discovery.