Question about small niche clusters when analyzing multiple Xenium samples with BANKSY

[Clement-zcj]

Hello, many thanks for developing BANKSY.

I am analyzing two Xenium slides with BANKSY, each contains 36 samples. When I perform niche identification, I notice that some clusters contain only a few dozen cells. Is this behavior reasonable?

I have tried reducing the number of dims used in RunUMAP and FindNeighbors, but the same phenomenon still occurs.

seu = RunBanksy(seu, lambda = 0.8, assay = 'RNA', slot = 'data',
dimx = 'x', dimy = 'y', features = 'all',
group = 'orig.ident', split.scale = TRUE, k_geom = 15)
seu = RunPCA(seu, assay = 'BANKSY', features = rownames(seu), npcs = 30)
seu <- readRDS('/work1/zcj/project/xenium/new/banksy/filter/niche.rds')
seu = RunUMAP(seu, dims = 1:20)
seu = FindNeighbors(seu, dims = 1:20)
seu = FindClusters(seu, resolution = 0.1)

head(seu@meta.data)
orig.ident nCount_RNA nFeature_RNA x y
aaaabeja-1_1 slice1_1D 89 78 35.49138 -110.57698
aaaacpbi-1_1 slice1_1D 82 75 29.30974 -88.87044
aaaadcig-1_1 slice1_1D 83 73 45.80009 -93.76973
aaaadgld-1_1 slice1_1D 23 21 50.21513 -98.47993
aaaaellb-1_1 slice1_1D 79 66 46.35295 -100.99507
aaaafbmm-1_1 slice1_1D 675 446 42.43059 -57.75594
cell_id staggered_sdimx staggered_sdimy BANKSY_snn_res.0.1
aaaabeja-1_1 aaaabeja-1_1 1804.008 -110.57698 3
aaaacpbi-1_1 aaaacpbi-1_1 1797.826 -88.87044 3
aaaadcig-1_1 aaaadcig-1_1 1814.317 -93.76973 3
aaaadgld-1_1 aaaadgld-1_1 1818.732 -98.47993 3
aaaaellb-1_1 aaaaellb-1_1 1814.870 -100.99507 3
aaaafbmm-1_1 aaaafbmm-1_1 1810.947 -57.75594 3
seurat_clusters sample
aaaabeja-1_1 3 slice1
aaaacpbi-1_1 3 slice1
aaaadcig-1_1 3 slice1
aaaadgld-1_1 3 slice1
aaaaellb-1_1 3 slice1
aaaafbmm-1_1 3 slice1

table(seu$BANKSY_snn_res.0.1)

0 1 10 11 12 13 14 2 3 4 5
958093 499510 765 469 256 25 20 300042 28561 17090 15249
6 7 8 9
1962 1668 910 789

─ Session info ─────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────
setting value
version R version 4.4.1 (2024-06-14)
os CentOS Linux 7 (Core)
system x86_64, linux-gnu
ui X11
language (EN)
collate en_US.UTF-8
ctype en_US.UTF-8
tz Asia/Shanghai
date 2025-08-14
pandoc NA

─ Packages ─────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────
! package * version date (UTC) lib source
P abind 1.4-8 2024-09-12 [5] CRAN (R 4.3.1)
aricode 1.0.3 2023-10-20 [5] CRAN (R 4.4.1)
Banksy * 1.5.6 2025-07-02 [5] Github (855c310)
VP Biobase 2.66.0 2021-10-26 [5] Bioconductor (on disk 2.54.0)
P BiocGenerics 0.52.0 2024-10-29 [5] Bioconductor 3.20 (R 4.4.1)
BiocManager 1.30.26 2025-06-05 [1] CRAN (R 4.4.1)
bitops 1.0-9 2024-10-03 [1] CRAN (R 4.3.3)
cli 3.6.5 2025-04-23 [1] CRAN (R 4.3.3)
cluster 2.1.2 2021-04-17 [2] CRAN (R 4.1.0)
codetools 0.2-20 2024-03-31 [1] CRAN (R 4.3.1)
colorspace 2.1-0 2023-01-23 [2] CRAN (R 4.1.0)
cowplot 1.1.3 2024-01-22 [2] CRAN (R 4.1.0)
P crayon 1.5.3 2024-06-20 [5] CRAN (R 4.3.1)
data.table 1.17.6 2025-06-17 [1] CRAN (R 4.3.3)
VP dbscan 1.2.2 2023-11-28 [5] CRAN (R 4.1.0) (on disk 1.1.12)
VP DelayedArray 0.32.0 2021-10-26 [5] Bioconductor (on disk 0.20.0)
P deldir 2.0-4 2024-02-28 [5] CRAN (R 4.3.3)
dichromat 2.0-0.1 2022-05-02 [2] CRAN (R 4.1.0)
P digest 0.6.37 2024-08-19 [5] CRAN (R 4.3.3)
P dotCall64 1.2 2024-10-04 [5] CRAN (R 4.3.3)
dplyr 1.1.4 2023-11-17 [2] CRAN (R 4.1.0)
farver 2.1.2 2024-05-13 [1] CRAN (R 4.3.1)
P fastDummies 1.7.5 2025-01-20 [5] CRAN (R 4.3.1)
P fastmap 1.2.0 2024-05-15 [5] CRAN (R 4.3.1)
P fitdistrplus 1.2-3 2025-07-01 [5] CRAN (R 4.3.3)
P future * 1.58.0 2025-06-05 [5] CRAN (R 4.3.3)
P future.apply 1.20.0 2025-06-06 [5] CRAN (R 4.3.3)
P generics 0.1.4 2025-05-09 [5] CRAN (R 4.3.3)
P GenomeInfoDb 1.42.3 2025-01-27 [5] Bioconductor 3.20 (R 4.4.1)
P GenomeInfoDbData 1.2.13 2025-07-01 [5] Bioconductor
GenomicRanges 1.46.1 2021-11-18 [2] Bioconductor
P ggplot2 * 3.5.2 2025-04-09 [5] CRAN (R 4.3.3)
P ggrepel 0.9.6 2024-09-07 [5] CRAN (R 4.3.3)
ggridges 0.5.6 2024-01-23 [2] CRAN (R 4.1.0)
P globals 0.18.0 2025-05-08 [5] CRAN (R 4.3.3)
P glue 1.8.0 2024-09-30 [5] CRAN (R 4.3.3)
goftest 1.2-3 2021-10-07 [2] CRAN (R 4.1.0)
gridExtra 2.3 2017-09-09 [2] CRAN (R 4.1.0)
P gtable 0.3.6 2024-10-25 [5] CRAN (R 4.3.1)
hms 1.1.3 2023-03-21 [2] CRAN (R 4.1.0)
P htmltools 0.5.8.1 2024-04-04 [5] CRAN (R 4.3.3)
htmlwidgets 1.6.4 2023-12-06 [2] CRAN (R 4.1.0)
httpuv 1.6.16 2025-04-16 [1] CRAN (R 4.3.3)
httr 1.4.7 2023-08-15 [2] CRAN (R 4.1.0)
ica 1.0-3 2022-07-08 [2] CRAN (R 4.1.0)
igraph 2.1.4 2025-01-23 [1] CRAN (R 4.3.1)
VP IRanges 2.40.1 2021-10-26 [5] Bioconductor (on disk 2.28.0)
P irlba 2.3.5.1 2022-10-03 [5] CRAN (R 4.3.1)
P jsonlite 2.0.0 2025-03-27 [5] CRAN (R 4.3.3)
KernSmooth 2.23-20 2021-05-03 [2] CRAN (R 4.1.0)
P later 1.4.2 2025-04-08 [5] CRAN (R 4.3.3)
lattice 0.22-5 2023-10-24 [2] CRAN (R 4.1.0)
lazyeval 0.2.2 2019-03-15 [2] CRAN (R 4.1.0)
P leidenAlg 1.1.5 2025-04-19 [5] CRAN (R 4.4.1)
P lifecycle 1.0.4 2023-11-07 [5] CRAN (R 4.1.0)
listenv 0.9.1 2024-01-29 [2] CRAN (R 4.1.0)
P lmtest 0.9-40 2022-03-21 [5] CRAN (R 4.3.3)
magick 2.8.3 2024-02-18 [2] CRAN (R 4.1.0)
magrittr 2.0.3 2022-03-30 [2] CRAN (R 4.1.0)
MASS 7.3-60.0.1 2024-01-13 [2] CRAN (R 4.1.0)
VP Matrix * 1.7-3 2023-11-30 [5] CRAN (R 4.3.3) (on disk 1.6.4)
MatrixGenerics 1.14.0 2023-10-24 [1] Bioconductor
P matrixStats 1.5.0 2025-01-07 [5] CRAN (R 4.3.3)
VP mclust 6.1.1 2021-12-17 [5] CRAN (R 4.1.0) (on disk 5.4.9)
P mime 0.13 2025-03-17 [5] CRAN (R 4.3.1)
P miniUI 0.1.2 2025-04-17 [5] CRAN (R 4.3.3)
nlme 3.1-155 2022-01-13 [2] CRAN (R 4.1.0)
P parallelly 1.45.0 2025-06-02 [5] CRAN (R 4.3.3)
P patchwork 1.3.1 2025-06-21 [5] CRAN (R 4.4.1)
pbapply 1.7-2 2023-06-27 [2] CRAN (R 4.1.0)
P pillar 1.10.2 2025-04-05 [5] CRAN (R 4.3.3)
pkgconfig 2.0.3 2019-09-22 [2] CRAN (R 4.1.0)
P plotly 4.11.0 2025-06-19 [5] CRAN (R 4.3.3)
plyr 1.8.9 2023-10-02 [1] CRAN (R 4.3.1)
png 0.1-8 2022-11-29 [2] CRAN (R 4.1.0)
P polyclip 1.10-7 2024-07-23 [5] CRAN (R 4.3.3)
P progressr 0.15.1 2024-11-22 [5] CRAN (R 4.3.3)
P promises 1.3.3 2025-05-29 [5] CRAN (R 4.4.1)
P purrr 1.0.4 2025-02-05 [5] CRAN (R 4.3.3)
R.methodsS3 1.8.2 2022-06-13 [2] CRAN (R 4.1.0)
R.oo 1.26.0 2024-01-24 [2] CRAN (R 4.1.0)
R.utils 2.12.3 2023-11-18 [2] CRAN (R 4.1.0)
P R6 2.6.1 2025-02-15 [5] CRAN (R 4.3.3)
P RANN 2.6.2 2024-08-25 [5] CRAN (R 4.3.1)
RColorBrewer 1.1-3 2022-04-03 [2] CRAN (R 4.1.0)
P Rcpp 1.0.14 2025-01-12 [5] CRAN (R 4.3.3)
RcppAnnoy 0.0.22 2024-01-23 [2] CRAN (R 4.1.0)
RcppHNSW 0.6.0 2024-02-04 [2] CRAN (R 4.1.0)
P RcppHungarian 0.3 2023-09-05 [5] CRAN (R 3.6.1)
RCurl 1.98-1.14 2024-01-09 [2] CRAN (R 4.1.0)
readr * 2.1.5 2024-01-10 [2] CRAN (R 4.1.0)
remotes 2.5.0 2024-03-17 [1] CRAN (R 4.3.1)
reshape2 1.4.4 2020-04-09 [2] CRAN (R 4.1.0)
P reticulate 1.42.0 2025-03-25 [5] CRAN (R 4.3.3)
VP rjson 0.2.23 2022-01-09 [5] CRAN (R 4.1.0) (on disk 0.2.21)
P rlang 1.1.6 2025-04-11 [5] CRAN (R 4.3.3)
ROCR 1.0-11 2020-05-02 [2] CRAN (R 4.1.0)
P RSpectra 0.16-2 2024-07-18 [5] CRAN (R 4.3.3)
rsvd 1.0.5 2021-04-16 [2] CRAN (R 4.1.0)
P Rtsne 0.17 2023-12-07 [5] CRAN (R 4.3.3)
P S4Arrays 1.6.0 2024-10-29 [5] Bioconductor 3.20 (R 4.4.1)
P S4Vectors 0.44.0 2024-10-29 [5] Bioconductor 3.20 (R 4.4.1)
P scales 1.4.0 2025-04-24 [5] CRAN (R 4.3.3)
scattermore 1.2 2023-06-12 [2] CRAN (R 4.1.0)
VP sccore 1.0.6 2021-12-12 [5] CRAN (R 4.1.0) (on disk 1.0.1)
P sctransform 0.4.2 2025-04-30 [5] CRAN (R 4.4.1)
sessioninfo * 1.2.2 2021-12-06 [2] CRAN (R 4.1.0)
P Seurat * 5.3.0 2025-04-23 [5] CRAN (R 4.3.3)
P SeuratObject * 5.1.0 2025-04-22 [5] CRAN (R 4.3.3)
P SeuratWrappers * 0.4.0 2025-08-05 [5] Github (jleechung/seurat-wrappers@392743e)
P shiny 1.11.0 2025-06-24 [5] CRAN (R 4.4.1)
P SingleCellExperiment 1.28.1 2024-11-10 [5] Bioconductor 3.20 (R 4.4.1)
P sp * 2.2-0 2025-02-01 [5] CRAN (R 4.3.3)
P spam 2.11-1 2025-01-20 [5] CRAN (R 4.3.3)
P SparseArray 1.6.2 2025-02-20 [5] Bioconductor 3.20 (R 4.4.1)
P SpatialExperiment 1.16.0 2024-10-29 [5] Bioconductor 3.20 (R 4.4.1)
P spatstat.data 3.1-6 2025-03-17 [5] CRAN (R 4.3.3)
spatstat.explore 3.4-3 2025-05-21 [1] CRAN (R 4.3.3)
spatstat.geom 3.4-1 2025-05-20 [1] CRAN (R 4.3.3)
spatstat.random 3.4-1 2025-05-20 [1] CRAN (R 4.3.3)
P spatstat.sparse 3.1-0 2024-06-21 [5] CRAN (R 4.3.3)
P spatstat.univar 3.1-3 2025-05-08 [5] CRAN (R 4.4.1)
P spatstat.utils 3.1-4 2025-05-15 [5] CRAN (R 4.4.1)
VP stringi 1.8.7 2023-12-11 [5] CRAN (R 4.1.0) (on disk 1.8.3)
stringr 1.5.1 2023-11-14 [2] CRAN (R 4.1.0)
SummarizedExperiment 1.24.0 2021-10-26 [2] Bioconductor
survival 3.5-8 2024-02-14 [2] CRAN (R 4.1.0)
P tensor 1.5.1 2025-06-17 [5] CRAN (R 4.3.3)
P tibble 3.3.0 2025-06-08 [5] CRAN (R 4.3.3)
tidyr 1.3.1 2024-01-24 [2] CRAN (R 4.1.0)
tidyselect 1.2.1 2024-03-11 [1] CRAN (R 4.3.1)
tzdb 0.4.0 2023-05-12 [2] CRAN (R 4.1.0)
P UCSC.utils 1.2.0 2024-10-29 [5] Bioconductor 3.20 (R 4.4.1)
P uwot 0.2.3 2025-02-24 [5] CRAN (R 4.3.3)
P vctrs 0.6.5 2023-12-01 [5] CRAN (R 4.1.0)
viridisLite 0.4.2 2023-05-02 [2] CRAN (R 4.1.0)
withr 3.0.2 2024-10-28 [1] CRAN (R 4.3.1)
xtable 1.8-4 2019-04-21 [2] CRAN (R 4.1.0)
XVector 0.34.0 2021-10-26 [2] Bioconductor
VP zlibbioc 1.52.0 2021-10-26 [5] Bioconductor (on disk 1.40.0)
P zoo 1.8-14 2025-04-10 [5] CRAN (R 4.3.3)

[jleechung]

Hi @Clement-zcj, thanks for using BANKSY!

Regarding small clusters: I'd first check if those small clusters are biological meaningful - you could find markers specific to that cluster (e.g. using FindAllMarkers), and see if you get anything meaningful there. It's plausible that these niches could correspond to rare cell types. Otherwise, this could just be due to noise in the data. It might be worth plotting the total transcript count or number of detected genes for these clusters and seeing if they're lower than the average in the data.

To mitigate the issue, you could also try increasing the k_geom parameter. This specifies the number of spatial neighbors to use when computing neighborhood features. Using larger values helps obtain smoother niches and may help avoid small fragmented clusters like what you're observing. In our domain segmentation vignette we use 30; you can play around with that and even try going up to 50.

As an aside - since you're dealing with two slides, it might also be possible that there are some batch effects present in the data, adding noise which the clustering picks up on. If increasing k_geom doesn't help, you can try using BANKSY with harmony (see here).