Abstract PR04: A network-based approach for drug synergy prediction from gene expression data

Abstract

Abstract The goal of combination therapy is to reduce the onset of drug resistance, while inducing a synergistic therapeutic effect, akin to synthetic lethality. Specifically, the measured reduction in tumor viability induced by a synergistic compound combination should be higher than expected based on additive behavior. Yet, assessing the synthetic lethality potential of a large collections of compounds is impractical, and the ability to predict drug synergy and synthetic lethality using quantitative approaches has not been experimentally validated. In addition, synthetic lethality has been shown to be cell-context dependent. To address these issues, we propose and experimentally validate a novel computational framework to efficiently explore the space of synthetic lethality in context specific fashion, to predict synergistic compound combinations. The reconstruction of accurate, context-specific molecular regulatory networks has shown significant potential in the rational elucidation of synergistic regulatory mechanisms in the cell. Our method leverages cell context-specific regulatory networks dissected by the ARACNe algorithm to infer changes in the activity of regulatory proteins (transcription factors and signaling proteins) based on the expression of their target genes. Specifically, we infer changes in protein activity, rather than in their corresponding gene expression, because most small molecule compounds modulate the activity but not the expression of their target proteins. We define a compound's functional Mode of Action (fMoA) as the specific protein activity signature induced by the compound in a specific cell type. To predict synergistic compounds, we propose two alternative approaches. (1) Target Molecular Phenotype (TMP): we first define a desired TMP to be induced by compound combination (e.g. cell death). We then leverage large collections of expression profiles following compound perturbations, such as the Connectivity Map (CMAP), to identify compounds whose combined fMoA most closely matches the TMP. We further prioritize compound pairs based on (a) the orthogonality of their individual fMoA, on the assumption that compounds with the same fMoA will affect the same pathways and thus will have a more additive behavior and (b) the complementary coverage of the TMP by their fMoA (i.e. compounds that implement distinct and complementary components of the TMP). (2) Sensitivity Induction (SI): we first obtain a protein activity signature of sensitivity to a given compound CA, by comparing the gene expression profiles of CA-sensitive vs CA-resistant cells. We then search for other compounds, whose fMoA most closely matches the CA's sensitivity signature, which are thus likely to act as sensitizers. We test the methods by using CMAP, as well as a dataset generated in our lab by profiling a set of three diffuse large B cell lymphoma (DLBCL) cell lines at 6h, 12h and 24h following perturbation with 92 drugs. We found that fMoA accurately captures the molecular targets of the assayed compounds. We then tested the synergistic predictions by measuring their overlap with all the pairs of 14 distinct compounds assessed in DLBCL cells (DREAM8 challenge data). For this analysis, we used both a general B-cell toxicity signature and an ABC-DLBCL-specific tumor progression signature as TMP. Finally, we studied the use of the SI approach to identify compounds that are candidate sensitizer of estrogen therapy (tamoxifen) in luminal ER+ breast tumors. This abstract is also presented as Poster A38. Citation Format: Mariano J. Alvarez, Yao Shen, Charles Karan, Mukesh Bansal, Michela Mattioli, Sergey Pampou, Andrea Califano. A network-based approach for drug synergy prediction from gene expression data. [abstract]. In: Proceedings of the AACR Precision Medicine Series: Synthetic Lethal Approaches to Cancer Vulnerabilities; May 17-20, 2013; Bellevue, WA. Philadelphia (PA): AACR; Mol Cancer Ther 2013;12(5 Suppl):Abstract nr PR04.