Elastic network modeling of cellular networks unveils sensor and effector genes that control information flow
Fig 1
Perturbation Response Scanning (PRS) applied to the yeast genetic interaction profile similarity network (GI PSN).
A) The PRS strategy illustrated on a toy network. The network is first transformed into a Laplacian matrix describing the network connectivity. The Laplacian matrix shows the degree of each node on the diagonal (shades of blue, darker is the higher degree) and connectivity of each node on the non-diagonal (red). Eigenvalue decomposition of the Laplacian yields the eigenvalues and eigenvectors used to calculate the covariance matrix. Each row of the covariance matrix is then normalized by the diagonal element resulting in an asymmetric PRS matrix where each row corresponds to a perturbed node, each column corresponds to a responding node and the colors show the magnitude of the response in arbitrary units (a.u., also used for the remaining figures). Row and column averages of the PRS matrix represent the effectiveness (right ordinate) and sensitivity (lower abscissa) of each node, respectively. B) PRS analysis of the GI PSN (left) yields the PRS matrix shown on the right. The nodes on the network (black dots) are the genes and the edges (gray lines) represent high profile similarity (Pearson’s correlation coefficient of pairs of genetic interaction profiles (PCC) ≥ 0.2). Gray shaded regions in the center of the network are the result of high connectivity and number of edges in the center. This network representation is used throughout the paper. Colored parts on the dendrograms along the two axes of the PRS matrix indicate distinct perturbed and responding gene clusters that are separated from the rest of the genes at the top level of the dendrograms (red and orange, respectively). These colors are used in the panels C and D to show the properties of the genes in these two clusters. C) Effectiveness (top) and sensitivity (bottom) boxplots showing the differences between perturbed and responding gene clusters, respectively (***: p < 0.001, permutation test). D) Representation of the distinct perturbed gene (top) and responding gene (bottom) clusters within the GI PSN.