Fig. 1

Overview of DeepLinc and the datasets. A Schematic description of the DeepLinc pipeline. DeepLinc consists of an encoder, a decoder, and an adversarial regularization module in the VGAE framework. The encoder is a two-layer graph convolutional network (GCN) and the decoder is a Sigmoid function for the dot product of latent variables. The single-cell transcriptome profile and the cell adjacency matrix derived from a single-cell spatial transcriptome dataset serve as inputs of DeepLinc. The output of DeepLinc is a new matrix (A’) presenting the reconstructed cell-cell interaction network. In addition, DeepLinc also uses the latent information of cell interaction landscapes and gene expression profiles for the visualization and clustering of single cells. B Statistics of the 4 datasets used for reconstruction of cell interaction networks by DeepLinc