<> "The repository administrator has not yet configured an RDF license."^^ . <> . . "An interaction-based modeling approach to predict response to cancer drugs"^^ . "In oncology, predictive biomarkers define patient subgroups that are likely to benefit from a specific cancer treatment. Since clinical studies entail high costs and low success rates, pre-clinical model systems like cancer cell lines are needed to generate biomarker hypotheses. Existing computational methods to predict drug response have several limitations. First, models often include large numbers of altered genes which contrasts with clinical predictive biomarkers that mostly include single altered genes. Second, models often assume that the effects of individual alterations are independent, although many biological processes rely on the interplay of multiple molecular components. \r\nWe developed an analytical framework to investigate the role of interactions in drug response based on linear regression models. Using data from two large cancer cell line panels, we conducted an exhaustive analysis of models with up to three genomic alterations. To increase model size, we constructed mutation interaction networks and applied module search algorithms to select subsets of mutations for drug response prediction models. We summarized important covariates as background models that served as a reference to evaluate the performance of models with genomic alterations. \r\nWe observed that including interactions increased the performance and robustness of drug response prediction models. Moreover, we identified several candidate interactions with consistent association patterns in two large cancer cell line panels. For example, we observed that cancer cell lines with BRAF and TP53 mutations showed worse response to BRAF inhibitors than cell lines with only BRAF mutations. Clinical data supports the resistance interaction between BRAF and TP53 mutations since patients with BRAF and TP53 mutations respond worse to the BRAF inhibitor Vemurafenib than patients with only BRAF mutations. This suggests that inhibition of the oncoprotein BRAF and reactivation of the tumor suppressor protein TP53 could be a promising combination therapy. Our analytical framework moreover allows to distinguish tissue-specific mutation associations from associations that are generalizable across tissues. In addition, we identified synthetic lethal triplets where the simultaneous mutation of two genes sensitizes cells to a drug. Our network-based approach outperformed a standard method for drug response prediction, the regularized regression algorithm elastic net. Based on 14 million models of different size, seven mutations were determined as the optimal model size. \r\nIn summary, we show that considering interactions in drug response prediction models unlocks a large predictive potential. Our interaction-based modeling approach contributes to a system-level understanding of the factors that mediate drug response."^^ . "2019" . . . . . . . "Dina Silvia"^^ . "Cramer"^^ . "Dina Silvia Cramer"^^ . . . . . . "An interaction-based modeling approach to predict response to cancer drugs (PDF)"^^ . . . "PhDThesis_CramerDina_20190325.pdf"^^ . . . "An interaction-based modeling approach to predict response to cancer drugs (Other)"^^ . . . . . . "indexcodes.txt"^^ . . . "An interaction-based modeling approach to predict response to cancer drugs (Other)"^^ . . . . . . "lightbox.jpg"^^ . . . "An interaction-based modeling approach to predict response to cancer drugs (Other)"^^ . . . . . . "preview.jpg"^^ . . . "An interaction-based modeling approach to predict response to cancer drugs (Other)"^^ . . . . . . "medium.jpg"^^ . . . "An interaction-based modeling approach to predict response to cancer drugs (Other)"^^ . . . . . . "small.jpg"^^ . . "HTML Summary of #28386 \n\nAn interaction-based modeling approach to predict response to cancer drugs\n\n" . "text/html" . . . "570 Biowissenschaften, Biologie"@de . "570 Life sciences"@en . .