To install the developmental version of the package, run:
To install from Bioconductor:
As complex tissues are typically composed of various cell types, deconvolution tools have been developed to computationally infer their cellular composition from bulk RNA sequencing (RNA-seq) data. To comprehensively assess deconvolution performance, gold-standard datasets are indispensable. Gold-standard, experimental techniques like flow cytometry or immunohistochemistry are resource-intensive and cannot be systematically applied to the numerous cell types and tissues profiled with high-throughput transcriptomics. The simulation of ‘pseudo-bulk’ data, generated by aggregating single-cell RNA-seq (scRNA-seq) expression profiles in pre-defined proportions, offers a scalable and cost-effective alternative. This makes it feasible to create in silico gold standards that allow fine-grained control of cell-type fractions not conceivable in an experimental setup. However, at present, no simulation software for generating pseudo-bulk RNA-seq data exists.
SimBu was developed to simulate pseudo-bulk samples based on various simulation scenarios, designed to test specific features of deconvolution methods. A unique feature of SimBu is the modelling of cell-type-specific mRNA bias using experimentally-derived or data-driven scaling factors. Here, we show that SimBu can generate realistic pseudo-bulk data, recapitulating the biological and statistical features of real RNA-seq data. Finally, we illustrate the impact of mRNA bias on the evaluation of deconvolution tools and provide recommendations for the selection of suitable methods for estimating mRNA content.
This chapter covers all you need to know to quickly simulate some pseudo-bulk samples!
This package can simulate samples from local or public data. This vignette will work with artificially generated data as it serves as an overview for the features implemented in SimBu. For the public data integration using sfaira (Fischer et al. 2020), please refer to the “Public Data Integration” vignette.
We will create some toy data to use for our simulations; two matrices with 300 cells each and 1000 genes/features. One represents raw count data, while the other matrix represents scaled TPM-like data. We will assign these cells to some immune cell types.
counts <- Matrix::Matrix(matrix(stats::rpois(3e5, 5), ncol = 300), sparse = TRUE)
tpm <- Matrix::Matrix(matrix(stats::rpois(3e5, 5), ncol = 300), sparse = TRUE)
tpm <- Matrix::t(1e6 * Matrix::t(tpm) / Matrix::colSums(tpm))
colnames(counts) <- paste0("cell_", rep(1:300))
colnames(tpm) <- paste0("cell_", rep(1:300))
rownames(counts) <- paste0("gene_", rep(1:1000))
rownames(tpm) <- paste0("gene_", rep(1:1000))
annotation <- data.frame(
"ID" = paste0("cell_", rep(1:300)),
"cell_type" = c(
rep("T cells CD4", 50),
rep("T cells CD8", 50),
rep("Macrophages", 100),
rep("NK cells", 10),
rep("B cells", 70),
rep("Monocytes", 20)
)
)
SimBu uses the SummarizedExperiment class as storage for count data as well as annotation data. Currently it is possible to store two matrices at the same time: raw counts and TPM-like data (this can also be some other scaled count matrix, such as RPKM, but we recommend to use TPMs). These two matrices have to have the same dimensions and have to contain the same genes and cells. Providing the raw count data is mandatory!
SimBu scales the matrix that is added via the tpm_matrix
slot by default to 1e6 per cell, if you do not want this, you can switch it off by setting the scale_tpm
parameter to FALSE
. Additionally, the cell type annotation of the cells has to be given in a dataframe, which has to include the two columns ID
and cell_type
. If additional columns from this annotation should be transferred to the dataset, simply give the names of them in the additional_cols
parameter.
To generate a dataset that can be used in SimBu, you can use the dataset()
method; other methods exist as well, which are covered in the “Inputs & Outputs” vignette.
ds <- SimBu::dataset(
annotation = annotation,
count_matrix = counts,
tpm_matrix = tpm,
name = "test_dataset"
)
#> Filtering genes...
#> Created dataset.
SimBu offers basic filtering options for your dataset, which you can apply during dataset generation:
filter_genes: if TRUE, genes which have expression values of 0 in all cells will be removed.
variance_cutoff: remove all genes with a expression variance below the chosen cutoff.
type_abundance_cutoff: remove all cells, which belong to a cell type that appears less the the given amount.
We are now ready to simulate the first pseudo bulk samples with the created dataset:
simulation <- SimBu::simulate_bulk(
data = ds,
scenario = "random",
scaling_factor = "NONE",
ncells = 100,
nsamples = 10,
BPPARAM = BiocParallel::MulticoreParam(workers = 4), # this will use 4 threads to run the simulation
run_parallel = TRUE
) # multi-threading to TRUE
#> Using parallel generation of simulations.
#> Warning in BiocParallel::MulticoreParam(workers = 4): MulticoreParam() not
#> supported on Windows, use SnowParam()
#> Finished simulation.
ncells
sets the number of cells in each sample, while nsamples
sets the total amount of simulated samples.
If you want to simulate a specific sequencing depth in your simulations, you can use the total_read_counts
parameter to do so. Note that this parameter is only applied on the counts matrix (if supplied), as TPMs will be scaled to 1e6 by default.
SimBu can add mRNA bias by using different scaling factors to the simulations using the scaling_factor
parameter. A detailed explanation can be found in the “Scaling factor” vignette.
Currently there are 6 scenarios
implemented in the package:
even: this creates samples, where all existing cell-types in the dataset appear in the same proportions. So using a dataset with 3 cell-types, this will simulate samples, where all cell-type fractions are 1/3. In order to still have a slight variation between cell type fractions, you can increase the balance_uniform_mirror_scenario
parameter (default to 0.01). Setting it to 0 will generate simulations with exactly the same cell type fractions.
random: this scenario will create random cell type fractions using all present types for each sample. The random sampling is based on the uniform distribution.
mirror_db: this scenario will mirror the exact fractions of cell types which are present in the provided dataset. If it consists of 20% T cells, 30% B cells and 50% NK cells, all simulated samples will mirror these fractions. Similar to the uniform scenario, you can add a small variation to these fractions with the balance_uniform_mirror_scenario
parameter.
weighted: here you need to set two additional parameters for the simulate_bulk()
function: weighted_cell_type
sets the cell-type you want to be over-representing and weighted_amount
sets the fraction of this cell-type. You could for example use B-cell
and 0.5
to create samples, where 50% are B-cells and the rest is filled randomly with other cell-types.
pure: this creates simulations of only one single cell-type. You have to provide the name of this cell-type with the pure_cell_type
parameter.
custom: here you are able to create your own set of cell-type fractions. When using this scenario, you additionally need to provide a dataframe in the custom_scenario_data
parameter, where each row represents one sample (therefore the number of rows need to match the nsamples
parameter). Each column has to represent one cell-type, which also occurs in the dataset and describes the fraction of this cell-type in a sample. The fractions per sample need to sum up to 1. An example can be seen here:
pure_scenario_dataframe <- data.frame(
"B cells" = c(0.2, 0.1, 0.5, 0.3),
"T cells" = c(0.3, 0.8, 0.2, 0.5),
"NK cells" = c(0.5, 0.1, 0.3, 0.2),
row.names = c("sample1", "sample2", "sample3", "sample4")
)
pure_scenario_dataframe
#> B.cells T.cells NK.cells
#> sample1 0.2 0.3 0.5
#> sample2 0.1 0.8 0.1
#> sample3 0.5 0.2 0.3
#> sample4 0.3 0.5 0.2
The simulation
object contains three named entries:
bulk
: a SummarizedExperiment object with the pseudo-bulk dataset(s) stored in the assays
. They can be accessed like this:utils::head(SummarizedExperiment::assays(simulation$bulk)[["bulk_counts"]])
#> 6 x 10 sparse Matrix of class "dgCMatrix"
#> [[ suppressing 10 column names 'random_sample1', 'random_sample2', 'random_sample3' ... ]]
#>
#> gene_1 513 497 470 504 504 493 533 496 496 521
#> gene_2 538 499 485 515 490 488 491 466 496 497
#> gene_3 515 454 482 444 536 433 453 468 464 497
#> gene_4 521 506 506 471 501 471 494 472 480 494
#> gene_5 544 494 493 491 482 471 518 541 473 493
#> gene_6 479 495 479 485 471 534 496 495 501 479
utils::head(SummarizedExperiment::assays(simulation$bulk)[["bulk_tpm"]])
#> 6 x 10 sparse Matrix of class "dgCMatrix"
#> [[ suppressing 10 column names 'random_sample1', 'random_sample2', 'random_sample3' ... ]]
#>
#> gene_1 1045.2238 990.6989 981.3188 941.2318 933.3490 940.4851 913.6014
#> gene_2 1029.3593 1032.8317 1073.5685 1041.7467 1066.9037 1048.0532 975.2949
#> gene_3 911.5577 1020.3936 971.2864 945.5291 1015.6197 909.1167 916.7610
#> gene_4 1084.0434 888.0826 960.9471 1043.4688 986.3028 937.6543 910.1941
#> gene_5 929.1642 1039.3336 969.9362 1010.1496 1002.1648 1141.7404 1008.7899
#> gene_6 1009.2036 936.5969 962.5341 1002.2696 952.2215 957.3881 945.0626
#>
#> gene_1 949.5001 896.0643 931.2061
#> gene_2 1033.1777 998.2513 1001.4575
#> gene_3 1065.3700 928.1785 1068.7896
#> gene_4 976.1890 919.9671 939.9508
#> gene_5 1077.4445 1047.4382 958.8432
#> gene_6 1035.7334 975.6425 926.5459
If only a single matrix was given to the dataset initially, only one assay is filled.
cell_fractions
: a table where rows represent the simulated samples and columns represent the different simulated cell-types. The entries in the table store the specific cell-type fraction per sample.
scaling_vector
: a named list, with the used scaling value for each cell from the single cell dataset.
It is also possible to merge simulations:
simulation2 <- SimBu::simulate_bulk(
data = ds,
scenario = "even",
scaling_factor = "NONE",
ncells = 1000,
nsamples = 10,
BPPARAM = BiocParallel::MulticoreParam(workers = 4),
run_parallel = TRUE
)
#> Using parallel generation of simulations.
#> Warning in BiocParallel::MulticoreParam(workers = 4): MulticoreParam() not
#> supported on Windows, use SnowParam()
#> Finished simulation.
merged_simulations <- SimBu::merge_simulations(list(simulation, simulation2))
Finally here is a barplot of the resulting simulation:
Sometimes, you are only interested in specific cell-types (for example T cells), but the dataset you are using has too many other cell-types; you can handle this issue during simulation using the whitelist
parameter:
simulation <- SimBu::simulate_bulk(
data = ds,
scenario = "random",
scaling_factor = "NONE",
ncells = 1000,
nsamples = 20,
BPPARAM = BiocParallel::MulticoreParam(workers = 4),
run_parallel = TRUE,
whitelist = c("T cells CD4", "T cells CD8")
)
#> Using parallel generation of simulations.
#> Warning in BiocParallel::MulticoreParam(workers = 4): MulticoreParam() not
#> supported on Windows, use SnowParam()
#> Finished simulation.
SimBu::plot_simulation(simulation = simulation)
In the same way, you can also provide a blacklist
parameter, where you name the cell-types you don’t want to be included in your simulation.
utils::sessionInfo()
#> R version 4.5.0 RC (2025-04-04 r88126 ucrt)
#> Platform: x86_64-w64-mingw32/x64
#> Running under: Windows Server 2022 x64 (build 20348)
#>
#> Matrix products: default
#> LAPACK version 3.12.1
#>
#> locale:
#> [1] LC_COLLATE=C
#> [2] LC_CTYPE=English_United States.utf8
#> [3] LC_MONETARY=English_United States.utf8
#> [4] LC_NUMERIC=C
#> [5] 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] SimBu_1.10.0
#>
#> loaded via a namespace (and not attached):
#> [1] tidyr_1.3.1 sass_0.4.10
#> [3] generics_0.1.3 SparseArray_1.8.0
#> [5] lattice_0.22-7 digest_0.6.37
#> [7] magrittr_2.0.3 RColorBrewer_1.1-3
#> [9] evaluate_1.0.3 sparseMatrixStats_1.20.0
#> [11] grid_4.5.0 fastmap_1.2.0
#> [13] jsonlite_2.0.0 Matrix_1.7-3
#> [15] GenomeInfoDb_1.44.0 proxyC_0.4.1
#> [17] httr_1.4.7 purrr_1.0.4
#> [19] scales_1.3.0 UCSC.utils_1.4.0
#> [21] codetools_0.2-20 jquerylib_0.1.4
#> [23] abind_1.4-8 cli_3.6.4
#> [25] rlang_1.1.6 crayon_1.5.3
#> [27] XVector_0.48.0 Biobase_2.68.0
#> [29] munsell_0.5.1 withr_3.0.2
#> [31] cachem_1.1.0 DelayedArray_0.34.0
#> [33] yaml_2.3.10 S4Arrays_1.8.0
#> [35] tools_4.5.0 parallel_4.5.0
#> [37] BiocParallel_1.42.0 dplyr_1.1.4
#> [39] colorspace_2.1-1 ggplot2_3.5.2
#> [41] GenomeInfoDbData_1.2.14 SummarizedExperiment_1.38.0
#> [43] BiocGenerics_0.54.0 vctrs_0.6.5
#> [45] R6_2.6.1 matrixStats_1.5.0
#> [47] stats4_4.5.0 lifecycle_1.0.4
#> [49] S4Vectors_0.46.0 IRanges_2.42.0
#> [51] pkgconfig_2.0.3 gtable_0.3.6
#> [53] bslib_0.9.0 pillar_1.10.2
#> [55] glue_1.8.0 data.table_1.17.0
#> [57] Rcpp_1.0.14 xfun_0.52
#> [59] tibble_3.2.1 GenomicRanges_1.60.0
#> [61] tidyselect_1.2.1 MatrixGenerics_1.20.0
#> [63] knitr_1.50 farver_2.1.2
#> [65] htmltools_0.5.8.1 labeling_0.4.3
#> [67] rmarkdown_2.29 compiler_4.5.0
Fischer, David S., Leander Dony, Martin König, Abdul Moeed, Luke Zappia, Sophie Tritschler, Olle Holmberg, Hananeh Aliee, and Fabian J. Theis. 2020. “Sfaira Accelerates Data and Model Reuse in Single Cell Genomics.” bioRxiv. https://doi.org/10.1101/2020.12.16.419036.