Abstract B2-41: Inference of differential dependency networks between tumor and stroma by integrated UNDO-KDDN software tool

Abstract

Abstract Introduction: Modeling biological networks is an important task in cancer systems biology research. Significant rewiring of molecular networks can drive key phenotypic transitions that can occur in both a tumor and its microenvironment. Tumor-stroma mixtures serve as both a major confounding factor and an underexploited information source in studying tumor systems. The co-evolution of neoplastic and stromal cells in a tumor can actively influence therapeutic response and shape resistance. Methods: We integrated unsupervised deconvolution into inference of tumor- or stroma-specific networks, for characterizing hidden crosstalk between tumor and stroma via network rewiring at hub genes. We developed a completely in silico unsupervised deconvolution method, namely UNDO, to dissect tumor-stroma mixed gene expressions in heterogeneous tumor samples. We formulated the inference of knowledge-fused differential dependency networks (KDDN) that incorporate both conditional data and prior knowledge as a convex optimization problem, and developed an efficient learning algorithm to jointly infer the conserved biological network and significant rewiring across different conditions. Results: We demonstrate the performance of UNDO over a wide range of tumor-stroma mixing proportions, validate UNDO on various biologically-mixed benchmark gene expression datasets, and further estimate tumor purity in TCGA/CPTAC datasets. The highly accurate deconvolution results obtained suggest not only the existence of cell-specific marker genes but also UNDO's ability to detect them blindly and correctly. Using KDDN, we detected that in utero estrogens induce a rewired network pattern in the mammary glands of rodent offspring that predicts for resistance to endocrine therapies. Subsequent studies have shown that tumors that arise in these mammary glands are less responsive to Tamoxifen (TAM) that represents the first study to explain why many ER+ breast cancers fail to respond (or respond and later recur) with TAM treatment. Moreover, focused on understanding how estrogen receptor-positive (ER+) breast cancer cells adapt to the stress of endocrine-based therapies, a gene network that coordinately regulates the functions of a cell that determine and execute the cell's fate decision, We followed the predictions of this topology and validated fundamentally new insights into molecular signaling, e.g., direct regulation of BCL2 by XBP1 and the requirement of NFκB for XBP1 signaling to regulate the prosurvival cell fate outcome in the cellular context of antiestrogen treatment and resistance. Conclusion: Tested on many benchmark datasets, UNDO is very effective at detecting cell-specific marker genes, estimating cell proportions and cell-specific expression profiles. Supported by a well-grounded mathematical framework, KDDN integrates the abundant biological knowledge and condition-specific experimental data to depict the overall dependency networks and their dynamics. We expect UNDO-KDDN to be a very useful tool for performing differential network analysis in many biological contexts. Citation Format: Lulu Chen, Niya Wang, Robert Clarke, Zhen Zhang, Yue Wang. Inference of differential dependency networks between tumor and stroma by integrated UNDO-KDDN software tool. [abstract]. In: Proceedings of the AACR Special Conference on Computational and Systems Biology of Cancer; Feb 8-11 2015; San Francisco, CA. Philadelphia (PA): AACR; Cancer Res 2015;75(22 Suppl 2):Abstract nr B2-41.