High-precision cell-type mapping and annotation of single-cell spatial transcriptomics with STAMapper.

In this paper, we develop a heterogeneous graph neural network, STAMapper, to transfer the cell-type labels from single-cell RNA-sequencing (scRNA-seq) data to single-cell spatial transcriptomics (scST) data. We collect 81 scST datasets consisting of 344 slices and 16 paired scRNA-seq datasets from eight technologies and five tissues to validate the efficiency of STAMapper. STAMapper achieves the best performance on 75 out of 81 datasets compared to competing methods in accuracy. STAMapper demonstrates enhanced performance over manual annotations, particularly at the boundaries of cell clusters, enables the unknown cell-type detection in scST data, and exhibits precise cell subtype annotations.