scpModel workflow/readSCP

[edemmott]

So, if I use the example data (Leduc_minimal), the modelling workflow works as described in the vignette. However when I try it with my own data it fails.

One difference is when I use readSCP on my own data, it seems to be generating a QFeatures object, whereas the Leduc_minimal dataset is recorded as Large SingleCellExperiment.

Running the standard SCoPE2 workflow (as described in that vignette) works on my data and doesnt mind it being described as a QFeatures rather than SCE object, it crashes at the point of using 'scpModelWorkflow', giving the error:

Error in scpModel(*tmp*, name) :
inherits(object, "SummarizedExperiment") is not TRUE

Environment:
R version 4.4.1 (2024-06-14)
Platform: aarch64-apple-darwin20
Running under: macOS Sonoma 14.5

Matrix products: default
BLAS: /System/Library/Frameworks/Accelerate.framework/Versions/A/Frameworks/vecLib.framework/Versions/A/libBLAS.dylib
LAPACK: /Library/Frameworks/R.framework/Versions/4.4-arm64/Resources/lib/libRlapack.dylib; LAPACK version 3.12.0

locale:
[1] en_US.UTF-8/en_US.UTF-8/en_US.UTF-8/C/en_US.UTF-8/en_US.UTF-8

time zone: Europe/London
tzcode source: internal

attached base packages:
[1] stats4 stats graphics grDevices utils datasets methods base

other attached packages:
[1] scpdata_1.13.0 ExperimentHub_2.13.1 AnnotationHub_3.13.3 BiocFileCache_2.13.0
[5] dbplyr_2.5.0 scran_1.33.2 scater_1.33.4 scuttle_1.15.4
[9] SingleCellExperiment_1.27.2 sva_3.53.0 BiocParallel_1.39.0 genefilter_1.87.0
[13] mgcv_1.9-1 nlme_3.1-166 impute_1.79.0 patchwork_1.2.0
[17] lubridate_1.9.3 forcats_1.0.0 stringr_1.5.1 dplyr_1.1.4
[21] purrr_1.0.2 readr_2.1.5 tidyr_1.3.1 tibble_3.2.1
[25] ggplot2_3.5.1 tidyverse_2.0.0 scp_1.15.1 BiocManager_1.30.25
[29] QFeatures_1.15.2 MultiAssayExperiment_1.31.5 SummarizedExperiment_1.35.1 Biobase_2.65.1
[33] GenomicRanges_1.57.1 GenomeInfoDb_1.41.1 IRanges_2.39.2 S4Vectors_0.43.2
[37] BiocGenerics_0.51.1 MatrixGenerics_1.17.0 matrixStats_1.4.1

loaded via a namespace (and not attached):
[1] RColorBrewer_1.1-3 rstudioapi_0.16.0 jsonlite_1.8.8 magrittr_2.0.3
[5] ggbeeswarm_0.7.2 farver_2.1.2 zlibbioc_1.51.1 vctrs_0.6.5
[9] memoise_2.0.1 BiocBaseUtils_1.7.3 S4Arrays_1.5.7 curl_5.2.2
[13] BiocNeighbors_1.99.0 SparseArray_1.5.34 plyr_1.8.9 cachem_1.1.0
[17] igraph_2.0.3 lifecycle_1.0.4 pkgconfig_2.0.3 rsvd_1.0.5
[21] Matrix_1.7-0 R6_2.5.1 fastmap_1.2.0 GenomeInfoDbData_1.2.12
[25] clue_0.3-65 fdrtool_1.2.18 colorspace_2.1-1 AnnotationDbi_1.67.0
[29] dqrng_0.4.1 irlba_2.3.5.1 lpsymphony_1.33.1 RSQLite_2.3.7
[33] beachmat_2.21.6 filelock_1.0.3 labeling_0.4.3 fansi_1.0.6
[37] timechange_0.3.0 httr_1.4.7 abind_1.4-8 compiler_4.4.1
[41] bit64_4.0.5 withr_3.0.1 viridis_0.6.5 DBI_1.2.3
[45] MASS_7.3-61 rappdirs_0.3.3 DelayedArray_0.31.11 bluster_1.15.1
[49] tools_4.4.1 vipor_0.4.7 beeswarm_0.4.0 glue_1.7.0
[53] grid_4.4.1 Rtsne_0.17 cluster_2.1.6 reshape2_1.4.4
[57] generics_0.1.3 gtable_0.3.5 tzdb_0.4.0 hms_1.1.3
[61] BiocSingular_1.21.3 ScaledMatrix_1.13.0 metapod_1.13.0 utf8_1.2.4
[65] XVector_0.45.0 BiocVersion_3.20.0 ggrepel_0.9.6 pillar_1.9.0
[69] limma_3.61.9 splines_4.4.1 lattice_0.22-6 survival_3.7-0
[73] bit_4.0.5 annotate_1.83.0 tidyselect_1.2.1 locfit_1.5-9.10
[77] Biostrings_2.73.1 gridExtra_2.3 ProtGenerics_1.37.1 edgeR_4.3.14
[81] IHW_1.33.0 statmod_1.5.0 stringi_1.8.4 UCSC.utils_1.1.0
[85] yaml_2.3.10 lazyeval_0.2.2 codetools_0.2-20 nipals_0.8
[89] MsCoreUtils_1.17.1 cli_3.6.3 xtable_1.8-4 munsell_0.5.1
[93] Rcpp_1.0.13 png_0.1-8 XML_3.99-0.17 parallel_4.4.1
[97] blob_1.2.4 AnnotationFilter_1.29.0 viridisLite_0.4.2 slam_0.1-53
[101] scales_1.3.0 crayon_1.5.3 rlang_1.1.4 cowplot_1.1.3
[105] KEGGREST_1.45.1

[edemmott]

Ok. So looking more at it, the QFeatures objects usually created contain the SCE as individual assays. Probably useful to change the example for the vignette over to the QFeatures object?

[lgatto]

One difference is when I use readSCP on my own data, it seems to be generating a QFeatures object, whereas the Leduc_minimal dataset is recorded as Large SingleCellExperiment.

All the data management from loading to QC, joining, aggregation, ... is indeed handled by QFeatures objects to keep track of all the steps. The modelling only requires the last set of the QFeatures object, that one can extract with getWithColData().

Probably useful to change the example for the vignette over to the QFeatures object?

Not sure I understand what you mean? readSCP() is a version that created a QFeatures object containing SingleCellExperiment sets. The readQFeatures() version documented in the QFeatures package returns a QFeatures object containing SummarizedExperiment sets.

[edemmott]

Sorry, I phrased badly. I think this is because the Leduc example dataset is the extracted single-cell experiment. So the vignette for the scpModel workflow starts from the singlecellexperiment leduc part. Wheres the more complete scp reproduction example using this dataset makes it clear you need to use getwithcoldata to extract the singlecellexperiment from the QFeatures object. The current vignete is just a little incomplete. Perhaps transfer over more of the scp repdocution example?

[lgatto]

Ok, thanks. That indeed makes sense, as some users might come straight to that vignette.

Instead of transferring more code (which leads to more maintenance and a longer, less straight-to-the-point vignette), I will add a sentence or to clarifying that we start immediately from the SingleCellExperiment, and point to the other vignette for more context.