Shin, D.-G. | bioRxiv Science

bioRxiv2024

vSPACE: Exploring Virtual Spatial Representation of Articular Chondrocytes at the Single-Cell Level

Single cell RNA sequencing technology has been dramatically changing how gene expression studies are performed. However, its use has been limited to identifying subtypes of cells by comparing cells gene expression levels in an unbiased manner to produce a 2D plot (e.g., UMAP/tSNE). We developed a new method of placing cells in 2D space. This system, called vSPACE, shows a virtual spatial representation of scRNAseq data obtained from human articular cartilage by emulating the concept of spatial transcriptomics technology, but virtually. This virtual 2D plot presentation of human articular cartage cells generates several zonal distribution patterns, in one or multiple genes at a time, reveling patterns that scientists can appreciate as imputed spatial distribution patterns along the zonal axis. The discovered patterns are explainable and remarkably consistent across all six healthy doners despite their respectively different clinical variables (age and sex), suggesting the confidence of the discovered patterns.

bioinformatics↗

bioRxiv2023

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

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.

bioinformatics↗