Systematic approaches for nominating combination therapies in cancer

Abstract

Combination therapy in cancer can provide enhanced anticancer efficacy and reduce the risk of drug resistance. Focused molecular experiments to discover combinations in a large search space are inefficient while high-throughput screening approaches do not provide sufficient mechanistic insights. We present two systematic approaches for nominating combination therapies, i) computational prediction of cellular response to unseen combination perturbations based on network models of cell biology, and ii) unbiased proteomic profiling of cellular response upon drug treatment to identify resistance mechanisms. Systematic perturbation of cells followed by comprehensive measurements of molecular and phenotypic responses provides information for constructing computational models of cell biology. Using a perturbation-response dataset of a melanoma cell line after drug treatments as a testbed, we developed a hybrid approach that combines explicit mathematical models of cell dynamics with a machine-learning framework to quantitatively predict cell behavior in response to perturbation of molecular targets. When sufficient perturbation-response data is available for model training, this model can predict cellular response to a vast number of unseen combination perturbations and therefore efficiently narrow down the search space and nominate promising sets of experimentally testable combination candidates. Investigation of resistance mechanisms can be used for the rational design of combination therapy. In an alternative approach, we used unbiased quantitative protein mass spectrometry to assess the cellular response profile to a small number of anti-cancer drug perturbations in ovarian cancer cells. Data-driven protein network analysis revealed known and novel markers of resistance, which we used to propose combination drug candidates. In a first round of validation experiments, synergistic and additive effects were observed for some combination candidates across multiple ovarian cancer cell lines, suggesting potential therapeutic value for future pre-clinical and clinical studies. Both systematic approaches can be used to effectively nominate combination therapies, provided the availability of sufficiently informative perturbation-response data. With further validation, the proposed combination candidates may contribute to the development of novel effective cancer therapeutics. We believe the approaches can be generalized to other cancer types and potentially be applied to other areas of cell biology.