Abstract:
|
Cell-cell communication is crucial for cellular activities. Spatial transcriptomics provides gene expression data in the context of spatial position. A few methods have been developed to infer cell-cell communication using spatial data, and spatial information has been shown to improve inference accuracy. Recently, there is an increasing number of spatial datasets that contain multiple replicates from different animals or tissue sections. Properly modeled, these replicates could have the potential for increasing confidence in downstream analysis. However, the existing methods for cell-cell communication inference do not easily incorporate replicates. Here, we develop a method using Gaussian processes to model cell-cell communications in spatial data with multiple samples. Our method models the spatial relationship of the expression between ligand and receptor genes, assuming a shared spatial correlation across samples. Our method does not require the alignment of spots or cells between samples, which currently is a challenging problem. By incorporating the data from multiple replicates, our method achieves higher accuracy in detecting communication signals than the existing method.
|