Translating melanoma molecular information into individualized protein network structures for the rational design of personalized treatment

Background: An urgent question in the frontline of melanoma research is: how can we assign the correct drugs to the correct patient? Large-scale datasets have been accumulating at an accelerating pace, and various statistical analyses are being developed aiming to extract significant signals from these massive data. However, it is still the case that tumors are assigned to pre-defined categories (BRAF mutant, NRAS mutant melanoma etc.), conceptually contradicting the vast heterogeneity that is known to exist among tumors, and likely overlooking unique tumors that need to be addressed and treated individually.

Methods: We use a thermodynamic-based, information-theoretical approach which allows us to decipher the structure of altered protein network in each tumor, that can comprise several altered subnetworks. Based on this analysis we predict and experimentally validate the response of the patient-specific network to different therapeutic modalities.

Results: Using a proteomic dataset, comprising 290 different cancer cell lines, including melanoma, we found that 17 altered protein subnetworks repeated themselves within the dataset. Each malignancy was characterized by a specific subset of 1-4 subnetworks out of 17. We validated experimentally that our predictions for the BRAF mutated melanoma worked at least as good as clinically prescribed treatments and in certain cases surpassed them.

Conclusion: The approach based on information theory, efficiently uncovers patient-specific altered proteomic network, allowing us to predict and rationally design smart, personalized drug combinations. Our method is highly selective: the predicted and very efficient drug combination for one melanoma is significantly less efficient for another and vice versa.