spSeudoMap: Cell type mapping of spatial transcriptomics using unmatched single-cell RNA-seq data

Abstract

With advances in computational models, the cellular landscape can be tracked in various tissues using spatial transcriptomics. Since many single-cell RNA-seq (scRNA-seq) data have been obtained after cell sorting, such as when investigating immune cells, integrating these singlecell data with spatial data is limited due to a mismatch of cell types composing the two datasets. Here, we present a method, spSeudoMap, which utilizes sorted scRNA-seq data to train a model for predicting cell types of spatial spots by creating virtual cell mixtures that closely mimic the gene expression profile of spatial transcriptomic data. To overcome the mismatch issue, the cell type exclusively present in the spatial data, pseudotype, was defined. The proportion of pseudotype cells and virtual expression profiles in the cell mixture was determined by pseudobulk transcriptomes. The simulated cell mixture was considered a reference dataset, and the model that predicts the cell composition of the mixture was trained to predict the cell fraction of the spatial data using domain adaptation. First, spSeudoMap was evaluated in human and mouse brain tissues, and the main region-specific neuron types extracted from single-cell data could be precisely mapped to the expected anatomical locations. Moreover, the method was applied to human breast cancer data and described the spatial distribution of immune cell subtypes and their interactions in heterogeneous tissue. Taken together, spSeudoMap is a platform that predicts the spatial composition of cell subpopulations using sorted scRNA-seq data, and it may help to clarify the roles of a few but crucial cell types.