Integration with single-cell RNA-seq

We use canonical correlation analysis (CCA) embedded in Seurat V3¹⁹ to integrate single-cell RNA-seq and single-cell ATAC-seq. We first calculate the gene accessibility account at variable genes identified using single-cell RNA-seq dataset. This can be done using a function called “createGmatFromMat” in SnapATAC package. Next, SnapATAC converts the snap object to a Seurat v3 object using a function called “SnapToSeurat” in preparation for integration. Different from integration method in Seurat, we use the low-dimension manifold as the dimensionality reduction method in the Seurat object. We next follow the vignette in Seurat website (https://satijalab.org/seurat/v3.0/atacseq_integration_vignette.html) to integrate these two modalities. The cell type for scATAC-seq is predicted using function “TransferData” in Seurat V3.

Finally, for each single-cell ATAC profile, we infer its gene expression profile by calculating the weighted average expression profile of its nearest neighboring cells in the single-cell RNA-seq dataset¹⁹. By doing so, we create pseudo-cells that contain information of both chromatin accessibility and gene expression profiles. The imputation of gene expression profile is done by “TransferData” function in Seurat V3.