Unexpected UMAP Structure in Pancreatic Visium HD Data Using Seurat v5

[mayurdoke6]

Dear Seurat Team,

Thank you for your continued development and support of Seurat v5. I am currently analyzing human pancreatic tissue data generated using the 10x Genomics Visium HD platform, and have encountered an issue with the UMAP visualization after running the standard spatial transcriptomics pipeline.

Despite following the recommended Seurat v5 workflow for spatial data (SCTransform, RunPCA, FindNeighbors, FindClusters, and RunUMAP with reduction = "pca"), the resulting UMAP appears highly compressed and poorly separated, without the expected clustering or biological structure. I have verified that:

The SCTransform normalization completed successfully.

PCA showed meaningful variance in the top components.

Clustering produced reasonable resolution.

Spatial plots look biologically plausible.

However, the UMAP does not reflect this structure and appears to collapse cells into a tight, poorly defined blob.

Questions:

1. Are there known issues or considerations when using UMAP on Visium HD data in Seurat v5?

2. Is there a recommended dimensionality reduction method (e.g., sketch PCA vs. standard PCA) or UMAP setting more appropriate for spatial HD data?

3. Could the high spot density or unique spatial resolution of Visium HD impact dimensionality reduction outcomes?

Any guidance or suggestions would be greatly appreciated. I am happy to provide code snippets or example data if needed.

Best regards,

Mayur

R version 4.4.1 (2024-06-14 ucrt) Platform: x86_64-w64-mingw32/x64 Running under: Windows 10 x64 (build 19045) Matrix products: default locale: [1] LC_COLLATE=English_United States.utf8 LC_CTYPE=English_United States.utf8 LC_MONETARY=English_United States.utf8 [4] LC_NUMERIC=C LC_TIME=English_United States.utf8 time zone: America/New_York tzcode source: internal attached base packages: [1] stats graphics grDevices utils datasets methods base other attached packages: [1] SeuratWrappers_0.4.0 tzdb_0.4.0 glmGamPoi_1.16.0 presto_1.0.0 data.table_1.16.4 [6] Rcpp_1.0.14 BPCells_0.3.0 patchwork_1.3.0 ggplot2_3.5.1 dplyr_1.1.4 [11] Seurat_5.2.1 SeuratObject_5.0.2 sp_2.1-4 loaded via a namespace (and not attached): [1] RColorBrewer_1.1-3 rstudioapi_0.17.1 jsonlite_1.8.9 magrittr_2.0.3 [5] spatstat.utils_3.1-2 ggbeeswarm_0.7.2 farver_2.1.2 zlibbioc_1.52.0 [9] vctrs_0.6.5 ROCR_1.0-11 spatstat.explore_3.3-4 S4Arrays_1.6.0 [13] htmltools_0.5.8.1 SparseArray_1.6.1 sctransform_0.4.1 parallelly_1.42.0 [17] KernSmooth_2.23-24 htmlwidgets_1.6.4 ica_1.0-3 plyr_1.8.9 [21] plotly_4.10.4 zoo_1.8-12 igraph_2.1.4 mime_0.12 [25] lifecycle_1.0.4 pkgconfig_2.0.3 rsvd_1.0.5 Matrix_1.7-1 [29] R6_2.6.1 fastmap_1.2.0 GenomeInfoDbData_1.2.13 MatrixGenerics_1.18.1 [33] fitdistrplus_1.2-2 future_1.34.0 shiny_1.10.0 digest_0.6.37 [37] colorspace_2.1-1 S4Vectors_0.44.0 tensor_1.5 RSpectra_0.16-2 [41] irlba_2.3.5.1 GenomicRanges_1.58.0 labeling_0.4.3 progressr_0.15.1 [45] spatstat.sparse_3.1-0 httr_1.4.7 polyclip_1.10-7 abind_1.4-8 [49] compiler_4.4.1 remotes_2.5.0 bit64_4.6.0-1 withr_3.0.2 [53] fastDummies_1.7.5 R.utils_2.12.3 MASS_7.3-60.2 DelayedArray_0.32.0 [57] tools_4.4.1 vipor_0.4.7 lmtest_0.9-40 beeswarm_0.4.0 [61] httpuv_1.6.15 future.apply_1.11.3 goftest_1.2-3 R.oo_1.27.0 [65] glue_1.8.0 nlme_3.1-164 promises_1.3.2 grid_4.4.1 [69] Rtsne_0.17 cluster_2.1.8 reshape2_1.4.4 generics_0.1.3 [73] hdf5r_1.3.12 gtable_0.3.6 spatstat.data_3.1-4 R.methodsS3_1.8.2 [77] tidyr_1.3.1 XVector_0.46.0 BiocGenerics_0.52.0 spatstat.geom_3.3-5 [81] RcppAnnoy_0.0.22 ggrepel_0.9.6 RANN_2.6.2 pillar_1.10.1 [85] stringr_1.5.1 spam_2.11-1 RcppHNSW_0.6.0 later_1.4.1 [89] splines_4.4.1 lattice_0.22-6 bit_4.5.0.1 survival_3.6-4 [93] deldir_2.0-4 tidyselect_1.2.1 miniUI_0.1.1.1 pbapply_1.7-2 [97] gridExtra_2.3 IRanges_2.40.1 SummarizedExperiment_1.36.0 scattermore_1.2 [101] stats4_4.4.1 Biobase_2.66.0 matrixStats_1.5.0 UCSC.utils_1.2.0 [105] stringi_1.8.4 lazyeval_0.2.2 codetools_0.2-20 tibble_3.2.1 [109] BiocManager_1.30.25 cli_3.6.3 uwot_0.2.2 arrow_18.1.0.1 [113] xtable_1.8-4 reticulate_1.40.0 munsell_0.5.1 GenomeInfoDb_1.42.3 [117] globals_0.16.3 spatstat.random_3.3-2 png_0.1-8 ggrastr_1.0.2 [121] spatstat.univar_3.1-1 parallel_4.4.1 assertthat_0.2.1 dotCall64_1.2 [125] listenv_0.9.1 viridisLite_0.4.2 scales_1.3.0 ggridges_0.5.6 [129] purrr_1.0.2 crayon_1.5.3 rlang_1.1.5 cowplot_1.1.3 --   > | > > 

[YuyanRuan]

Hi, I got similar results due to low-UMI data. I got it looks better by reducing the number of PCs. I set dims=c(1:15) in my analysis (the tutorial used 50)