Fig. 3

Evaluating MISTy on mechanistic in silico data. A MISTy was evaluated on the task of reconstruction of simulated interaction networks. Models of intra- and intercellular interactions of four different cell types (cell type specific intracellular networks are shown in Sup. Fig. 2), arranged on a grid representing a tissue, were used to simulate measurements of 29 molecular species. We considered two pipelines, (1) in which cell type information is available and (2) where cell types are not considered. B Increase in explained variance by adding the paraview contribution to the intraview model. Only variables with positive paraview contribution are shown. C Contribution of each view to the prediction of the marker expressions in the meta-model. The stacked barplot represents normalized values of the fusion coefficients of the respective views for each marker. D Receiver operating characteristic (ROC) depicting the aggregate performance of MISTy on the samples for the intraview and paraview, for the two cases with and without cell type information. The dashed lines represent the expected performance of an uninformed classifier, the gray iso-lines represent points in ROC space with informedness (Youden’s J statistic) equal to 0.1, 0.2, 0.5, and 0.8. E Predicted importance of the interactions for the intraview and paraview models for the case with cell type information (for cell type 1) together with the direct interactions from the in silico model (red crosses). Some targets had very low variance and therefore filtered out