Resolution:
## Figure 5.
Plate diagrams of the two Interaction Component Models. (A) Plain ICM. Each interaction is assumed to have been generated from a component z, by sampling the end points i and j independently from the multinomial distribution ϕ_{z}. (B) Extension (ICMg2) with gene expression data x_{i }in the nodes; the lower part of the plate is the same as in (A). Node data x_{i }are assumed to have been generated from the same components z as the interactions, by sampling each node i in the data from ϕ_{z }and then its gene expression profile from the normal distribution N (μ_{z}, σ^{2}I). Symbols: multinomial distribution θ over components, component-specific multinomial distributions ϕ_{z }over nodes, distribution hyperparameters α and β, prior and component-specific means μ_{0 }and μ_{z}, prior and shared noise standard deviations σ_{0 }and σ, number of components C, links N and nodes M.
Parkkinen and Kaski |