A correlation-based network model of signaling is inferred using the SkMel-133 perturbation response data and the graphical Gaussian model (GGM) network inference method. GGM uses the first-order partial correlation coefficients as a measure of direct associations between two nodes, for which the effects of other nodes are eliminated. The advantage of partial correlations over the conventional pairwise correlations is its ability to distinguish direct interactions from indirect ones. Partial coefficients are calculated by inverting the correlation matrix. Here, we computed the partial correlations using the shrinkage estimation method, a form of regularized estimator of the covariance matrix and its inverse (Schafer and Strimmer, 2005). Shrinkage method is particularly useful for inferring models when number of variables is higher than number of constraints (n > p). See (Schafer and Strimmer, 2005) for algorithmic details and formulation of partial correlation calculation. The network model covers the interactions between 138 proteomic species and five phenotypic measurements (i.e., cell viability and cell cycle progression nodes). The models involve the interactions with statistically significant (False Discovery Rate (FDR) -adjusted p-value < 0.01, H0: pcor = 0 based on a null-distribution estimated from the data, see Schafer and Strimmer, 2005, for details) association strengths (green: positive, red: negative associations). The resulting models capture some of the well-characterized and potentially interesting interactions such as those between AKT, GSK3, and TSC2. Cell viability node is directly linked to proteomic nodes SMAD3, Cyclin E1, MGMT. G1-arrest node is directly linked to MEK phosphorylation node, while S-arrest is associated with beta-catenin phosphorylation node. G2-arrest is linked to STAT3 and caveolin total levels, and G2M transition is linked to BRAF, RB, and p90S6K total level nodes. The models provide an approximate picture of pairwise interactions in SkMel-133 cells, are not executable, do not capture the high-order couplings within the cellular system, and hence, we conclude that they do not have the required power for predicting system response to untested perturbations.