Since read counts are summed across cells in a pseudobulk approach, modeling continuous cell-level covariates also requires a collapsing step. Here we summarize the values of a variable from a set of cells using the mean, and store the value for each cell type. Including these variables in a regression formula uses the summarized values from the corresponding cell type.
We demonstrate this feature on a lightly modified analysis of PBMCs from 8 individuals stimulated with interferon-β (Kang, et al, 2018, Nature Biotech).
Here is the code from the main vignette:
library(dreamlet)
library(muscat)
library(ExperimentHub)
library(scater)
# Download data, specifying EH2259 for the Kang, et al study
eh <- ExperimentHub()
sce <- eh[["EH2259"]]
# only keep singlet cells with sufficient reads
sce <- sce[rowSums(counts(sce) > 0) > 0, ]
sce <- sce[, colData(sce)$multiplets == "singlet"]
# compute QC metrics
qc <- perCellQCMetrics(sce)
# remove cells with few or many detected genes
ol <- isOutlier(metric = qc$detected, nmads = 2, log = TRUE)
sce <- sce[, !ol]
# set variable indicating stimulated (stim) or control (ctrl)
sce$StimStatus <- sce$stim
In many datasets, continuous cell-level variables could be mapped reads, gene count, mitochondrial rate, etc. There are no continuous cell-level variables in this dataset, so we can simulate two from a normal distribution:
sce$value1 <- rnorm(ncol(sce))
sce$value2 <- rnorm(ncol(sce))
Now compute the pseudobulk using standard code:
sce$id <- paste0(sce$StimStatus, sce$ind)
# Create pseudobulk
pb <- aggregateToPseudoBulk(sce,
assay = "counts",
cluster_id = "cell",
sample_id = "id",
verbose = FALSE
)
The means per variable, cell type, and sample are stored in the pseudobulk SingleCellExperiment
object:
metadata(pb)$aggr_means
## # A tibble: 128 × 5
## # Groups: cell [8]
## cell id cluster value1 value2
## <fct> <fct> <dbl> <dbl> <dbl>
## 1 B cells ctrl101 3.96 0.0500 -0.109
## 2 B cells ctrl1015 4.00 -0.0559 -0.0398
## 3 B cells ctrl1016 4 -0.00683 0.143
## 4 B cells ctrl1039 4.04 0.0563 -0.144
## 5 B cells ctrl107 4 -0.123 -0.0438
## 6 B cells ctrl1244 4 -0.0337 -0.0223
## 7 B cells ctrl1256 4.01 -0.00792 0.0519
## 8 B cells ctrl1488 4.02 -0.0139 0.00624
## 9 B cells stim101 4.09 0.145 0.216
## 10 B cells stim1015 4.06 -0.0949 -0.0345
## # ℹ 118 more rows
Including these variables in a regression formula uses the summarized values from the corresponding cell type. This happens behind the scenes, so the user doesn’t need to distinguish bewteen sample-level variables stored in colData(pb)
and cell-level variables stored in metadata(pb)$aggr_means
.
Variance partition and hypothesis testing proceeds as ususal:
form <- ~ StimStatus + value1 + value2
# Normalize and apply voom/voomWithDreamWeights
res.proc <- processAssays(pb, form, min.count = 5)
# run variance partitioning analysis
vp.lst <- fitVarPart(res.proc, form)
# Summarize variance fractions genome-wide for each cell type
plotVarPart(vp.lst, label.angle = 60)
# Differential expression analysis within each assay
res.dl <- dreamlet(res.proc, form)
# dreamlet results include coefficients for value1 and value2
res.dl
## class: dreamletResult
## assays(8): B cells CD14+ Monocytes ... Megakaryocytes NK cells
## Genes:
## min: 164
## max: 5262
## details(7): assay n_retain ... n_errors error_initial
## coefNames(4): (Intercept) StimStatusstim value1 value2
A variable in colData(sce)
is handled according to if the variable is
metadata(pb)$aggr_means
colData(pb)
## R version 4.5.0 Patched (2025-04-21 r88169)
## Platform: aarch64-apple-darwin20
## Running under: macOS Ventura 13.7.1
##
## Matrix products: default
## BLAS: /Library/Frameworks/R.framework/Versions/4.5-arm64/Resources/lib/libRblas.0.dylib
## LAPACK: /Library/Frameworks/R.framework/Versions/4.5-arm64/Resources/lib/libRlapack.dylib; LAPACK version 3.12.1
##
## locale:
## [1] C/en_US.UTF-8/en_US.UTF-8/C/en_US.UTF-8/en_US.UTF-8
##
## time zone: America/New_York
## tzcode source: internal
##
## attached base packages:
## [1] stats4 stats graphics grDevices utils datasets methods
## [8] base
##
## other attached packages:
## [1] muscData_1.23.0 scater_1.37.0
## [3] scuttle_1.19.0 ExperimentHub_2.99.5
## [5] AnnotationHub_3.99.5 BiocFileCache_2.99.5
## [7] dbplyr_2.5.0 muscat_1.23.0
## [9] dreamlet_1.7.1 SingleCellExperiment_1.31.0
## [11] SummarizedExperiment_1.39.0 Biobase_2.69.0
## [13] GenomicRanges_1.61.0 GenomeInfoDb_1.45.4
## [15] IRanges_2.43.0 S4Vectors_0.47.0
## [17] BiocGenerics_0.55.0 generics_0.1.4
## [19] MatrixGenerics_1.21.0 matrixStats_1.5.0
## [21] variancePartition_1.39.0 BiocParallel_1.43.2
## [23] limma_3.65.1 ggplot2_3.5.2
## [25] BiocStyle_2.37.0
##
## loaded via a namespace (and not attached):
## [1] bitops_1.0-9 httr_1.4.7
## [3] RColorBrewer_1.1-3 doParallel_1.0.17
## [5] Rgraphviz_2.53.0 numDeriv_2016.8-1.1
## [7] tools_4.5.0 sctransform_0.4.2
## [9] backports_1.5.0 utf8_1.2.5
## [11] R6_2.6.1 metafor_4.8-0
## [13] mgcv_1.9-3 GetoptLong_1.0.5
## [15] withr_3.0.2 prettyunits_1.2.0
## [17] gridExtra_2.3 cli_3.6.5
## [19] sandwich_3.1-1 labeling_0.4.3
## [21] sass_0.4.10 KEGGgraph_1.69.0
## [23] SQUAREM_2021.1 mvtnorm_1.3-3
## [25] blme_1.0-6 mixsqp_0.3-54
## [27] zenith_1.11.0 dichromat_2.0-0.1
## [29] parallelly_1.44.0 invgamma_1.1
## [31] RSQLite_2.3.11 shape_1.4.6.1
## [33] gtools_3.9.5 dplyr_1.1.4
## [35] Matrix_1.7-3 metadat_1.4-0
## [37] ggbeeswarm_0.7.2 abind_1.4-8
## [39] lifecycle_1.0.4 multcomp_1.4-28
## [41] yaml_2.3.10 edgeR_4.7.2
## [43] mathjaxr_1.8-0 gplots_3.2.0
## [45] SparseArray_1.9.0 grid_4.5.0
## [47] blob_1.2.4 crayon_1.5.3
## [49] lattice_0.22-7 beachmat_2.25.0
## [51] msigdbr_24.1.0 annotate_1.87.0
## [53] KEGGREST_1.49.0 magick_2.8.6
## [55] pillar_1.10.2 knitr_1.50
## [57] ComplexHeatmap_2.25.0 rjson_0.2.23
## [59] boot_1.3-31 estimability_1.5.1
## [61] corpcor_1.6.10 future.apply_1.11.3
## [63] codetools_0.2-20 glue_1.8.0
## [65] data.table_1.17.4 vctrs_0.6.5
## [67] png_0.1-8 Rdpack_2.6.4
## [69] gtable_0.3.6 assertthat_0.2.1
## [71] cachem_1.1.0 zigg_0.0.2
## [73] xfun_0.52 rbibutils_2.3
## [75] S4Arrays_1.9.1 Rfast_2.1.5.1
## [77] coda_0.19-4.1 reformulas_0.4.1
## [79] survival_3.8-3 iterators_1.0.14
## [81] tinytex_0.57 statmod_1.5.0
## [83] TH.data_1.1-3 nlme_3.1-168
## [85] pbkrtest_0.5.4 bit64_4.6.0-1
## [87] filelock_1.0.3 progress_1.2.3
## [89] EnvStats_3.1.0 bslib_0.9.0
## [91] TMB_1.9.17 irlba_2.3.5.1
## [93] vipor_0.4.7 KernSmooth_2.23-26
## [95] colorspace_2.1-1 rmeta_3.0
## [97] DBI_1.2.3 DESeq2_1.49.1
## [99] tidyselect_1.2.1 emmeans_1.11.1
## [101] bit_4.6.0 compiler_4.5.0
## [103] curl_6.2.3 httr2_1.1.2
## [105] graph_1.87.0 BiocNeighbors_2.3.1
## [107] DelayedArray_0.35.1 bookdown_0.43
## [109] scales_1.4.0 caTools_1.18.3
## [111] remaCor_0.0.18 rappdirs_0.3.3
## [113] stringr_1.5.1 digest_0.6.37
## [115] minqa_1.2.8 rmarkdown_2.29
## [117] aod_1.3.3 XVector_0.49.0
## [119] RhpcBLASctl_0.23-42 htmltools_0.5.8.1
## [121] pkgconfig_2.0.3 lme4_1.1-37
## [123] sparseMatrixStats_1.21.0 mashr_0.2.79
## [125] fastmap_1.2.0 rlang_1.1.6
## [127] GlobalOptions_0.1.2 UCSC.utils_1.5.0
## [129] DelayedMatrixStats_1.31.0 farver_2.1.2
## [131] jquerylib_0.1.4 zoo_1.8-14
## [133] jsonlite_2.0.0 BiocSingular_1.25.0
## [135] RCurl_1.98-1.17 magrittr_2.0.3
## [137] Rcpp_1.0.14 babelgene_22.9
## [139] viridis_0.6.5 EnrichmentBrowser_2.39.0
## [141] stringi_1.8.7 MASS_7.3-65
## [143] plyr_1.8.9 listenv_0.9.1
## [145] parallel_4.5.0 ggrepel_0.9.6
## [147] Biostrings_2.77.1 splines_4.5.0
## [149] hms_1.1.3 circlize_0.4.16
## [151] locfit_1.5-9.12 reshape2_1.4.4
## [153] ScaledMatrix_1.17.0 BiocVersion_3.22.0
## [155] XML_3.99-0.18 evaluate_1.0.3
## [157] RcppParallel_5.1.10 BiocManager_1.30.25
## [159] nloptr_2.2.1 foreach_1.5.2
## [161] tidyr_1.3.1 purrr_1.0.4
## [163] future_1.49.0 clue_0.3-66
## [165] scattermore_1.2 ashr_2.2-63
## [167] rsvd_1.0.5 broom_1.0.8
## [169] xtable_1.8-4 fANCOVA_0.6-1
## [171] viridisLite_0.4.2 truncnorm_1.0-9
## [173] tibble_3.2.1 lmerTest_3.1-3
## [175] glmmTMB_1.1.11 memoise_2.0.1
## [177] beeswarm_0.4.0 AnnotationDbi_1.71.0
## [179] cluster_2.1.8.1 globals_0.18.0
## [181] GSEABase_1.71.0