Batch-effect correction

Batch integration was initially evaluated for menstrual and secretory samples separately using 3 methods including Fast mutual nearest neighbors (fastMNN) (in package batchelor (version 1.2.4)) [21], Harmony (version 1.0) [22], and Seurat v3 [19, 20] recommended by two benchmark studies [23, 24]. Briefly, data normalization and HVGs defining were performed firstly for each batch (donor). FastMNN was applied using RunFastMNN function of SeuratWrapper package by setting the order as true. Harmony was applied using RunHarmony function of SeuratWrapper package after PCA dimensional reduction. For Seurat v3, anchors across batches were identified using the FindIntegrationAnchors and the data were finally integrated using the IntegrateData function of the Seurat package. The assay “integrated” was used for downstream analysis including data scaling, dimensional reduction and clustering. For secretory and clonogenic cells, batch effect was removed using Harmony given its superiority in integrating samples either with shared subpopulation(s) or with distinct cell types.