Batch correction using Harmony

[abbaslab]

On https://portals.broadinstitute.org/harmony/SeuratV3.html, the recommended batch correction steps are:
pbmc <- pbmc %>%
RunUMAP(reduction = "harmony", dims = 1:20) %>%
FindNeighbors(reduction = "harmony", dims = 1:20) %>%
FindClusters(resolution = 0.5) %>%
identity()

Its a bit confusing since the RunUMAP function is going first, then FindNeighbors and FindClusters. In other Seurat vignettes, RunUMAP is ran last.

Is there a specific recommendation? I tried both ways and surprisingly, the RunUMAP then FindNighbors then FindClusters had more biologically meaningful separation of few clusters. Is that expected? Does it matter what order one is using?

[timoast]

This is a vignette from a different group, not the Seurat developers, but the order of running UMAP and clustering will not matter.