To combine cells together across samples, we normalized the UMI counts with the variance-stabilizing SCTransform and glmGamPoi on each sample135,136 and jointly embedded cells across samples with reciprocal PCA integration134, as outlined in https://satijalab.org/seurat/articles/integration_rpca.html. In this joint embedding, we over-clustered the dataset with FindClusters(algorithm = 2, resolution = 1)and removed any cluster with more than 10% of cells flagged by miQC, scds, or DropletQC as low-quality biased-clusters in the data.
