Cell signaling characterization for spatial transcriptomics (ST) data using network analysis

We describe a network analysis-based cell-cell communication method for Spatial Transcriptomics (ST) data. For each evaluated ligand-receptor interaction, we define a fully connected, directed and weighted network model where nodes represent the individual ST locations with directed edge weights set to the product of the reduced rank reconstructed expression values for the ligand at the source location and cognate receptor at the target location divided by the squared distance between the locations. Using this network, we compute the weighted in-degree centrality to quantify signaling activity of the target ligand-receptor interaction at each location. Our method is validated for three interactions on a real ST dataset against five different cell-cell communication strategies. We report that our method captures the simultaneous expression heterogeneity in both the ligand and the receptor and generates biologically plausible cell communication profiles for the Wnt3-Fzd1, Ephb1-Efnb3 and Ptprc-Cd22 interactions. An important finding of this work is the importance of building network models for ST data using a low dimensional embedding of the gene-level data.