Learning cell communication from spatial graphs of cells - Node centric expression models (NCEM)
Abstract:
Spatial molecular profiling assays have enabled single-cell genomics to move from studying tissue heterogeneity to tissue organization, providing new avenues to study cellular communication within a tissue. Yet, with the new information comes additional complexity: deriving insights from this data requires a new set of analysis tools. We present a computational method based on graph neural networks which reconciles disentanglement of gene expression variation and cell communication modeling: node-centric expression modeling (NCEM). The NCEM approach is not limited to targeted spatial transcriptomics data, but can be extended to spot transcriptomics if within-cell-type variation can be recovered in deconvolution analyses. The statistical cell-cell dependencies discovered by NCEM are plausible signatures of known molecular processes underlying cell communication. We show that NCEM's cell type coupling analysis workflow and the identification of receiver and sender effects across multiple datasets finds plausible putative dependencies and niche-dependent cell state variation on the example of human lymph nodes, inflamed colon and colorectal cancer in both targeted and spot transcriptomics data.
13 сен 2024