Fig. 1
Overview of the INSPIRE framework. INSPIRE takes as input multiple expression datasets that potentially contain different sets of genes and learns a network of expression modules (i.e. co-expressed sets of genes) conserved across these datasets. INSPIRE is a general framework that can take any number of datasets as input; two datasets (X 1 and X 2 ) are shown in representation for simplicity. Top left: Two input datasets are represented by rectangles with black solid lines. Rows represent genes and columns represent samples. The blue region contains the data for the genes that are contained in both datasets. The pink and green regions contain the data for the genes which are contained by only one of the datasets. Top right: The features (latent variables), each corresponding to a module, are shown by the orange matrix as learned by INSPIRE. These are used as an LDR of the expression datasets. Top middle: As an example, five INSPIRE features L 1, …, L 5 (orange-shaded circles), 12 genes G 1, …, G 12 associated with those features, and the conditional dependency network among the INSPIRE features are represented. The dependencies among features are conserved across the datasets. Bottom: Five modules, each corresponding to an INSPIRE feature, and the dependency network among them are represented as the interpretation of the INSPIRE features and their conditional dependencies