When dealing with multiple samples in RDS, should we opt for reduced dimensionality using 'HARMONY' or stick with PCA?

[hai178912522]

sce <- as.SingleCellExperiment(data) class: SingleCellExperiment dim: 15341 79286 metadata(0): assays(2): counts logcounts rownames(15341): SAMD11 NOC2L ... PPAN-P2RY11 CEACAM19 rowData names(0): colnames(79286): P30.post.1_P30.tr.1.AAACCTGAGATCCGAG-1 P30.post.1_P30.tr.1.AAACCTGAGCCGATTT-1 ... P28.pre_P28.ut.TTTGTCAGTGACCAAG-1 P28.pre_P28.ut.TTTGTCATCGAATGCT-1 colData names(13): orig.ident nCount_RNA ... seurat_cluster ident reducedDimNames(4): UMAP TSNE HARMONY PCA mainExpName: RNA altExpNames(0): milo <- Milo(sce) milo <- buildGraph(milo, k = 30, d = 30, reduced.dim = "HARMONY") milo <- makeNhoods(milo, prop = 0.05, k = 30, d = 30, refined = TRUE, reduced_dims = "HARMONY") p.NhSize = plotNhoodSizeHist(milo) milo <- countCells(milo, meta.data = as.data.frame(colData(milo)), sample="sample") 6 x 30 sparse Matrix of class "dgCMatrix"

1 2 1 3 2 2 1 1 7 2 . . 2 2 1 1 11 3 8 . 6 5 8 5 . 2 1 1 . 1 1 2 2 3 . 1 1 . 2 1 1 1 . . . 1 . 8 . 4 . 3 6 4 11 . . . . . . . 3 3 1 3 2 . 1 1 2 1 1 . . 1 4 3 14 . 6 2 5 6 10 16 1 1 1 1 . 3 . 4 . 12 5 3 . 5 7 4 2 . . 4 6 1 2 20 4 11 2 6 5 6 10 . 4 2 8 . 1 1 5 4 3 3 1 1 1 2 7 . 3 . 2 3 1 3 14 1 8 3 9 14 7 6 . 4 . 2 . . . 6 1 7 1 1 1 2 5 6 1 1 . . 6 2 3 20 3 8 . 4 5 7 16 1 4 4 1 . 2 .

Design

design <- data.frame(colData(milo))[,c("sample", "group")] design <- distinct(design) rownames(design) <- design$sample

Compute neighbourhood connectivity

milo <- calcNhoodDistance(milo, d=30, reduced.dim = "HARMONY")

Differential abundance testing

da_results <- testNhoods(milo, design = ~ group, design.df = design) ggplot(da_results, aes(PValue)) + geom_histogram(bins=50)

ggplot(da_results, aes(logFC, -log10(SpatialFDR))) + geom_point() + geom_hline(yintercept = 1) ## Ma

Is there any issue with my analysis? Will it result in lack of significance?

[emdann]

Hi @hai178912522, It makes perfect sense to use the Harmony dimensionality reduction for constructing the KNN graph, if this is the integration methods you have been using for other graph based analyses on your dataset (e.g. clustering, UMAP). This looks like a fairly standard case where no significant differences in abundance are detected, probably because of a small number of replicates per condition (3 replicates as far as I can tell).

One caveat with using Harmony is that it tends to overcorrect differences between samples, as opposed to other integration methods, so it might be that differences between conditions are lost after integration.

[hai178912522]

This is the situation with my design. It is a comparison of 8 vs 22 in total。 A data.frame: 30 × 2 sample group

P30.post.1 P30.post.1 Post-treatment P2.pre P2.pre Treatment naive P29.pre P29.pre Treatment naive P29.post.1 P29.post.1 Post-treatment P37.post.1 P37.post.1 Post-treatment P8.pre P8.pre Treatment naive P38.post.1 P38.post.1 Post-treatment P26.pre P26.pre Treatment naive P14.pre P14.pre Treatment naive P9.pre P9.pre Treatment naive P7.pre P7.pre Treatment naive P4.pre P4.pre Treatment naive P15.pre P15.pre Treatment naive P27.pre P27.pre Treatment naive P22.pre P22.pre Treatment naive P1.post.3 P1.post.3 Post-treatment P33.post.1 P33.post.1 Post-treatment P12.pre P12.pre Treatment naive P36.post.1 P36.post.1 Post-treatment P34.pre P34.pre Treatment naive P6.pre P6.pre Treatment naive P23.pre P23.pre Treatment naive P33.pre P33.pre Treatment naive P20.pre P20.pre Treatment naive P35.post.1 P35.post.1 Post-treatment P17.pre P17.pre Treatment naive P30.pre P30.pre Treatment naive P18.pre P18.pre Treatment naive P35.pre P35.pre Treatment naive P28.pre P28.pre Treatment naive Is there any good method to help adjust the results?

[MikeDMorgan]

Closing due to inactivity