Abstract:Cell-cell communication is crucial in maintaining cellular homeostasis, cell survival and various regulatory relationships among interacting cells. Thanks to recent advances of spatial transcriptomics technologies, we can now explore if and how cells' proximal information available from spatial transcriptomics datasets can be used to infer cell-cell communication. Here we present a cell-cell communication inference framework, called CGCom, which uses a graph neural network (GNN) to learn communication patterns among interacting cells by combining single-cell spatial transcriptomic datasets with publicly available ligand-receptor information and the molecular regulatory information down-stream of the ligand-receptor signaling. To evaluate the performance of CGCom, we applied it to mouse embryo seqFISH datasets. Our results demonstrate that CGCom can not only accurately infer cell communication between individual cell pairs but also generalize its learning to predict communication between different cell types. We compared the performance of CGCom with two existing methods, CellChat and CellPhoneDB, and our comparative study revealed both common and unique communication patterns from the three approaches. Commonly found communication patterns include three sets of ligand-receptor communication relationships, one between surface ectoderm cells and spinal cord cells, one between gut tube cells and endothelium, and one between neural crest and endothelium, all of which have already been reported in the literature thus offering credibility of all three methods. However, we hypothesize that CGCom is superior in reducing false positives thanks to its use of cell proximal information and its learning between specific cell pairs rather than between cell types. CGCom is a GNN-based solution that can take advantage of spatially resolved single-cell transcriptomic data in predicting cell-cell communication with a higher accuracy.

GCNG: graph convolutional networks for inferring gene interaction from spatial transcriptomics data

STdGCN: spatial transcriptomic cell-type deconvolution using graph convolutional networks

Spatial-MGCN: a novel multi-view graph convolutional network for identifying spatial domains with attention mechanism

HyperGCN: an effective deep representation learning framework for the integrative analysis of spatial transcriptomics data

SpatialcoGCN: deconvolution and spatial information–aware simulation of spatial transcriptomics data via deep graph co-embedding

Identifying spatial domains of spatially resolved transcriptomics via multi-view graph convolutional networks

Inferring single-cell resolution spatial gene expression via fusing spot-based spatial transcriptomics, location and histology using GCN

SpaGCN: Integrating gene expression, spatial location and histology to identify spatial domains and spatially variable genes by graph convolutional network

Revealing cell–cell communication pathways with their spatially coupled gene programs

Cross-view graph neural networks for spatial domain identification by integrating gene expression, spatial locations with histological images

ICN: extracting interconnected communities in gene co-expression networks

Gene Expression Prediction Based on Neighbour Connection Neural Network Utilizing Gene Interaction Graphs.

A comprehensive overview of graph neural network-based approaches to clustering for spatial transcriptomics

scGCN is a graph convolutional networks algorithm for knowledge transfer in single cell omics.

MGEGFP: a multi-view graph embedding method for gene function prediction based on adaptive estimation with GCN

TransGCN: a semi-supervised graph convolution network–based framework to infer protein translocations in spatio-temporal proteomics

GMFGRN: a matrix factorization and graph neural network approach for gene regulatory network inference

CLARIFY: cell–cell interaction and gene regulatory network refinement from spatially resolved transcriptomics

CGCom: a framework for inferring Cell-cell Communication based on Graph Neural Network

Cell Clustering for Spatial Transcriptomics Data with Graph Neural Networks

VCNet: Vector-Based Gene Co-Expression Network Construction and Its Application to RNA-seq Data.