effect, Harmony69 integration was performed on Seurat’s log-normalized and scaled data while regressing out cell cycle scores (Supplementary Fig. 5a). In this step, a set of genes that lacked biological relevance to our research and displayed uneven expression patterns (Supplementary Table 1), thereby exerting an impact on the clustering results, were removed from the dataset. Seurat’s graph-based clustering and UMAP visualization based on the first 30 Harmony components resulted in identifying contaminating clusters of T cells and macrophages (Supplementary Fig. 5b). These clusters were discarded, and the workflow of integration and clustering (resolution = 0.5) was repeated. To perform differential gene expression, each dataset was processed individually using the “SCTransform” (v2) workflow70 and “PrepSCTFindMarkers” was applied to the SCT assays to eliminate the impact of fluctuating sequencing depths. To identify the markers of each cluster, Seurat’s “FindConservedMarker” was used, and genes with adjusted p-value < 0.05 in all conditions were considered significant. The overall log fold change (LFC) was calculated by averaging the LFC over conditions for each gene.
