PMC:4307189 / 9054-13288 JSONTXT

{"target":"https://pubannotation.org/docs/sourcedb/PMC/sourceid/4307189","sourcedb":"PMC","sourceid":"4307189","source_url":"https://www.ncbi.nlm.nih.gov/pmc/4307189","text":"In this study, we collected the gene expression values of the ErbB2-positive parental SKBR3 cell-line and the lapatinib-resistant SKBR3-R cell-line, derived from it, in the presence and absence of lapatinib [17]. Then we used a fully Bayesian statistical modeling approach to identify and analyze characteristic drug-resistant cross-talks between EGFR/ErbB and other signaling pathways. ln that process, we considered two gene-gene networks originating from the gene expression matrices of both parental and resistant conditions, individually. To say a gene-pair involved in cross-talk between two particular signaling pathways has high potential of being involved in acquired drug-resistance, our research hypothesis was it should have high probability of appearing in the resistant network and low probability in the parental network. The rationale behind our hypothesis was that in breast cancer cell lines resistant to tamoxifen, a cross-talk mechanism has previously been identified between EGFR and the IGF1R signaling pathway [18]. The schematic diagram of our proposed framework is shown in Figure 1. Like other biological processes, cancer signaling pathway activities and their corresponding network data possess stochasticity such that some gene-gene relationships (i.e. network edges) may not always be present or detected, whereas some other typical relationships may be absent. The stochastic nature of biological systems can be used to predict edge probabilities by formalizing them into a probabilistic model with other network properties [19]. Hill et al. reported a data-driven approach that exploits a Dynamic Bayesian Network (DBN) model to infer probabilistic relationships between node-pairs in a context-specific signaling network [20]. This study incorporates existing signaling biology using an informative prior distribution on the network, and its weight of contribution is measured with an empirical Bayes analysis, maximum marginal likelihood. This study predicts a number of known and unexpected signaling links through time that are validated using independent targeted inhibition experiments [20]. Here we have used a fully Bayesian approach for inferring a probabilistic model: a special class of Exponential Random Graph Model, namely the p1-model. We used Gibbs sampling for estimating model parameters with non-informative priors, in order to estimate the posterior probabilities of edges in gene-gene relationship networks. These identified cross-talks do not appear in the parental network but only in the resistant one, because the signaling network can be ‘rewired’ in a specific context [21,22]. This idea resembles the approach taken by Hill et al. in that they inferred the probabilities of signaling links (gene-pairs) varying through time. Thus, these drug-resistance cross-talks can be informative to elucidate the complex mechanisms underlying drug-insensitivity and can help to develop novel therapeutics targeting signaling pathways.\nFigure 1 Schematic diagram of our proposed framework. (A) The framework for finding putative drug-resistant cross-talks. At first two gene expression data matrices were generated individually from the samples of both parental and resistant conditions. Next, based on pair-wise correlations of genes’ expression values, two gene-gene relationship networks were derived. Then, a Bayesian statistical model called the p 1-model was applied on those two networks to find posterior probabilities of network edges. These posterior probabilities were used to find gene-pairs potentially contributing to drug resistance. Next, these gene-pairs were analyzed for overlap with cross-talks between EGFR/ErbB and other signaling pathways, and thus putative drug-resistant cross-talks were identified. (B) Hierarchical Bayesian model for inferring posterior probabilities of network parameters. Here, α represents the propensity (expansiveness/attractiveness) of a gene to be connected in an undirected network, and is dependent on the hyperparameter Σ; θ is the global density parameter; λ ij=l o g(n ij) is the scaling parameter, which is fixed due to the constraint ∑kYijk=1; the hyperparameter τ θ represents precision of the normal prior for the parameter θ.","divisions":[{"label":"label","span":{"begin":2983,"end":2991}}],"tracks":[{"project":"2_test","denotations":[{"id":"25599599-16037379-14868115","span":{"begin":1034,"end":1036},"obj":"16037379"},{"id":"25599599-20100321-14868116","span":{"begin":1556,"end":1558},"obj":"20100321"},{"id":"25599599-22923301-14868117","span":{"begin":1755,"end":1757},"obj":"22923301"},{"id":"25599599-22923301-14868118","span":{"begin":2125,"end":2127},"obj":"22923301"},{"id":"25599599-22579283-14868119","span":{"begin":2629,"end":2631},"obj":"22579283"},{"id":"25599599-17322911-14868120","span":{"begin":2632,"end":2634},"obj":"17322911"}],"attributes":[{"subj":"25599599-16037379-14868115","pred":"source","obj":"2_test"},{"subj":"25599599-20100321-14868116","pred":"source","obj":"2_test"},{"subj":"25599599-22923301-14868117","pred":"source","obj":"2_test"},{"subj":"25599599-22923301-14868118","pred":"source","obj":"2_test"},{"subj":"25599599-22579283-14868119","pred":"source","obj":"2_test"},{"subj":"25599599-17322911-14868120","pred":"source","obj":"2_test"}]}],"config":{"attribute types":[{"pred":"source","value type":"selection","values":[{"id":"2_test","color":"#e2ec93","default":true}]}]}}