After the above quality control, all cells from the Seurat objects for each pool were integrated into the same Seurat object for visualization in the same 2D space. The atomic sketch integration method was used, a dictionary learning based procedure recently developed in Seurat for large datasets (see https://satijalab.org/seurat/articles/parsebio_sketch_integration). Briefly, 5,000 representative cells were selected from each pool (based on statistical leverage). Integration was performed on these sketched cells using the reference-based RPCAIntegration method. Then, each cell from each pool was placed in this integrated space as well using the ProjectIntegration function.
