Decipher cell communication with attention: CLARA

Cell-cell communication is crucial in maintaining cellular homeostasis, cell survival and various regulatory relationships among interacting cells. Thanks to recent advances in spatial transcriptomics technologies, we can now explore how cells proximal information can be used to infer cell-cell communication. Most current studies on cell communication focus on reporting general interactions between cell types without cell proximal information. While few methods detailing communication for each cell pair, two main draw backs can be found in these methods. I. These cells has fail to provide a reasonable cell communication radius and depend on pre-define. II. These methods tend to find cell communication based on highly expression genes which are inferred by comparing global gene expression values instead of local expression value. This will lead to miss identifying local high expression values. Here we present a cell-cell communication inference framework, called GACom, which converts a cell communication problem into a natural language process problem and infers each cell pairs communication based on the cell-transformed word relationship with an altered attention mechanism named ligand-receptor attention. We used one mouse embryo SeqFISH dataset and one human cartilage dataset to demonstrate the performance of CLARA. In Mouse embryo dataset, we explore the communication pattern through Vascular endothelial growth factor. In human cartilage dataset, we explore how the DKK1[->]LRP6 and IL6[->]IL6ST change at normal and osteoarthritis. We also demonstrate how the CLARA can provide a new view for cell sub grouping including the cells that are sending the ligand and the cell that are receiving ligand. The biological process analysis shows the rationality of the grouping strategy. CLARA is an attention-based cell communication inference solution that leverages spatially or virtually spatially resolved single-cell transcriptomic data to predict cell-cell communication, further providing a strategic grouping approach based on cell communication insights.