A Network Polypharmacological Approach to Combinatorial Drug Repurposing for Diffuse Intrinsic Pontine Glioma

Despite five decades of clinical investigations, there is currently no effective treatment for children diagnosed with Diffuse Intrinsic Pontine Glioma (DIPG). We now understand that DIPGs share the same histone 3 mutation and fatal prognosis as other diffuse midline gliomas (DMGs), which led to the introduction of a new entity referred to as DMG, H3 K27M mutant. Indeed, therapeutics indicated for other brain neoplasms have proven ineffective for DIPGs. We posit that by using a polypharmacological approach to determine drug combinations that target distinct mechanistic pathways of DIPG, it is more likely that an efficacious treatment will be developed. We predict monodrug therapies using a link prediction model trained on various embeddings of a drug-disease regulatory network and physicochemical properties of small molecules and proteins. We validate the in silico predictions by performing cell viability assays on patient-derived cell cultures for notable therapeutics. Using FDA-approved drugs as a proxy for viability of a drug pair for combinatorial use, we develop a model to predict the synergism of the relationship between drug pairs. Finally, we calculate the transitive probability that a drug pair contains drugs that individually regulate DIPG, are blood-brain barrier penetrant, and the drug pair are suitable for combined use. We find only moderate agreement between computational predictions and experimental results for both monodrug and multidrug therapies, we believe due to the heterogeneity of the disease, the difficulties of modelling brain permeability, and an inherent literature bias in the knowledge graph. Such challenges need to be overcome to develop an efficacious therapy for this disease.