library(devil)
library(scRNAseq)
#> Loading required package: SingleCellExperiment
#> Loading required package: SummarizedExperiment
#> Loading required package: MatrixGenerics
#> Loading required package: matrixStats
#>
#> Attaching package: 'MatrixGenerics'
#> The following objects are masked from 'package:matrixStats':
#>
#> colAlls, colAnyNAs, colAnys, colAvgsPerRowSet, colCollapse,
#> colCounts, colCummaxs, colCummins, colCumprods, colCumsums,
#> colDiffs, colIQRDiffs, colIQRs, colLogSumExps, colMadDiffs,
#> colMads, colMaxs, colMeans2, colMedians, colMins, colOrderStats,
#> colProds, colQuantiles, colRanges, colRanks, colSdDiffs, colSds,
#> colSums2, colTabulates, colVarDiffs, colVars, colWeightedMads,
#> colWeightedMeans, colWeightedMedians, colWeightedSds,
#> colWeightedVars, rowAlls, rowAnyNAs, rowAnys, rowAvgsPerColSet,
#> rowCollapse, rowCounts, rowCummaxs, rowCummins, rowCumprods,
#> rowCumsums, rowDiffs, rowIQRDiffs, rowIQRs, rowLogSumExps,
#> rowMadDiffs, rowMads, rowMaxs, rowMeans2, rowMedians, rowMins,
#> rowOrderStats, rowProds, rowQuantiles, rowRanges, rowRanks,
#> rowSdDiffs, rowSds, rowSums2, rowTabulates, rowVarDiffs, rowVars,
#> rowWeightedMads, rowWeightedMeans, rowWeightedMedians,
#> rowWeightedSds, rowWeightedVars
#> Loading required package: GenomicRanges
#> Loading required package: stats4
#> Loading required package: BiocGenerics
#>
#> Attaching package: 'BiocGenerics'
#> The following objects are masked from 'package:stats':
#>
#> IQR, mad, sd, var, xtabs
#> The following objects are masked from 'package:base':
#>
#> anyDuplicated, aperm, append, as.data.frame, basename, cbind,
#> colnames, dirname, do.call, duplicated, eval, evalq, Filter, Find,
#> get, grep, grepl, intersect, is.unsorted, lapply, Map, mapply,
#> match, mget, order, paste, pmax, pmax.int, pmin, pmin.int,
#> Position, rank, rbind, Reduce, rownames, sapply, setdiff, table,
#> tapply, union, unique, unsplit, which.max, which.min
#> Loading required package: S4Vectors
#>
#> Attaching package: 'S4Vectors'
#> The following object is masked from 'package:utils':
#>
#> findMatches
#> The following objects are masked from 'package:base':
#>
#> expand.grid, I, unname
#> Loading required package: IRanges
#> Loading required package: GenomeInfoDb
#> Loading required package: Biobase
#> Welcome to Bioconductor
#>
#> Vignettes contain introductory material; view with
#> 'browseVignettes()'. To cite Bioconductor, see
#> 'citation("Biobase")', and for packages 'citation("pkgname")'.
#>
#> Attaching package: 'Biobase'
#> The following object is masked from 'package:MatrixGenerics':
#>
#> rowMedians
#> The following objects are masked from 'package:matrixStats':
#>
#> anyMissing, rowMedians
#> Warning: replacing previous import 'S4Arrays::makeNindexFromArrayViewport' by
#> 'DelayedArray::makeNindexFromArrayViewport' when loading 'SummarizedExperiment'
#> Warning: replacing previous import 'S4Arrays::makeNindexFromArrayViewport' by
#> 'DelayedArray::makeNindexFromArrayViewport' when loading 'HDF5Array'This vignette provides a step-by-step guide to using the devil package for differential expression analysis of single-cell RNA sequencing (scRNA-seq) data. We’ll start by loading a dataset, preparing the data, fitting the model, and then testing for differential expression between conditions. Finally, we’ll visualize the results using a volcano plot.
Input data
First, we load a single-cell RNA dataset from the
scRNAseq package. This dataset contains RNA counts for
individual cells and metadata describing the features of each cell.
data <- scRNAseq::ReprocessedFluidigmData()Extracting Counts and Metadata
To fit a model using devil, we need to extract the RNA
counts and the associated metadata.
counts <- data@assays@data[[1]]
metadata <- data@colData
print(dim(counts)) # Dimensions of the counts matrix
#> [1] 26255 130
print(dim(metadata)) # Dimensions of the metadata
#> [1] 130 28
print(head(metadata)) # Display the first few rows of metadata
#> DataFrame with 6 rows and 28 columns
#> NREADS NALIGNED RALIGN TOTAL_DUP PRIMER INSERT_SZ
#> <numeric> <numeric> <numeric> <numeric> <numeric> <numeric>
#> SRR1275356 10554900 7555880 71.5862 58.4931 0.0217638 208
#> SRR1274090 196162 182494 93.0323 14.5122 0.0366826 247
#> SRR1275251 8524470 5858130 68.7213 65.0428 0.0351827 230
#> SRR1275287 7229920 5891540 81.4884 49.7609 0.0208685 222
#> SRR1275364 5403640 4482910 82.9609 66.5788 0.0298284 228
#> SRR1275269 10729700 7806230 72.7536 50.4285 0.0204349 245
#> INSERT_SZ_STD COMPLEXITY NDUPR PCT_RIBOSOMAL_BASES
#> <numeric> <numeric> <numeric> <numeric>
#> SRR1275356 63 0.868928 0.343113 2e-06
#> SRR1274090 133 0.997655 0.935730 0e+00
#> SRR1275251 89 0.789252 0.201082 0e+00
#> SRR1275287 78 0.898100 0.538191 0e+00
#> SRR1275364 76 0.890693 0.391660 0e+00
#> SRR1275269 99 0.879414 0.431169 0e+00
#> PCT_CODING_BASES PCT_UTR_BASES PCT_INTRONIC_BASES
#> <numeric> <numeric> <numeric>
#> SRR1275356 0.125806 0.180954 0.613229
#> SRR1274090 0.309822 0.412917 0.205185
#> SRR1275251 0.398461 0.473884 0.039886
#> SRR1275287 0.196420 0.227592 0.498944
#> SRR1275364 0.138617 0.210406 0.543941
#> SRR1275269 0.333077 0.354635 0.248331
#> PCT_INTERGENIC_BASES PCT_MRNA_BASES MEDIAN_CV_COVERAGE
#> <numeric> <numeric> <numeric>
#> SRR1275356 0.080008 0.306760 1.495770
#> SRR1274090 0.072076 0.722739 1.007580
#> SRR1275251 0.087770 0.872345 1.242990
#> SRR1275287 0.077044 0.424013 0.775981
#> SRR1275364 0.107035 0.349024 1.441370
#> SRR1275269 0.063957 0.687712 0.617100
#> MEDIAN_5PRIME_BIAS MEDIAN_3PRIME_BIAS MEDIAN_5PRIME_TO_3PRIME_BIAS
#> <numeric> <numeric> <numeric>
#> SRR1275356 0.000000 0.166122 1.036250
#> SRR1274090 0.181742 0.698991 0.293510
#> SRR1275251 0.000000 0.340046 0.201518
#> SRR1275287 0.010251 0.350915 0.292838
#> SRR1275364 0.000000 0.204074 0.619863
#> SRR1275269 0.057960 0.345502 0.284480
#> sample_id.x Lane_ID LibraryName avgLength spots
#> <character> <character> <character> <integer> <integer>
#> SRR1275356 SRX534610 D24VYACXX130502:4 GW16_2 202 9818076
#> SRR1274090 SRX534823 1 NPC_9 60 95454
#> SRR1275251 SRX534623 D24VYACXX130502:4 GW16_8 202 7935952
#> SRR1275287 SRX534641 D24VYACXX130502:1 GW21+3_2 202 6531944
#> SRR1275364 SRX534614 D24VYACXX130502:7 GW16_23 202 4919561
#> SRR1275269 SRX534632 D24VYACXX130502:4 GW21_8 202 9969377
#> Biological_Condition Coverage_Type Cluster1 Cluster2
#> <character> <character> <factor> <factor>
#> SRR1275356 GW16 High IIIb III
#> SRR1274090 NPC Low 1a I
#> SRR1275251 GW16 High NA III
#> SRR1275287 GW21+3 High 1c I
#> SRR1275364 GW16 High IIIb III
#> SRR1275269 GW21 High NA I
colnames(metadata) # Display the column names of metadata
#> [1] "NREADS" "NALIGNED"
#> [3] "RALIGN" "TOTAL_DUP"
#> [5] "PRIMER" "INSERT_SZ"
#> [7] "INSERT_SZ_STD" "COMPLEXITY"
#> [9] "NDUPR" "PCT_RIBOSOMAL_BASES"
#> [11] "PCT_CODING_BASES" "PCT_UTR_BASES"
#> [13] "PCT_INTRONIC_BASES" "PCT_INTERGENIC_BASES"
#> [15] "PCT_MRNA_BASES" "MEDIAN_CV_COVERAGE"
#> [17] "MEDIAN_5PRIME_BIAS" "MEDIAN_3PRIME_BIAS"
#> [19] "MEDIAN_5PRIME_TO_3PRIME_BIAS" "sample_id.x"
#> [21] "Lane_ID" "LibraryName"
#> [23] "avgLength" "spots"
#> [25] "Biological_Condition" "Coverage_Type"
#> [27] "Cluster1" "Cluster2"The output shows that we have 130 cells described by 28 features, along with RNA counts for 26,255 genes.
Preprocessing: Filtering Non-Expressed Genes
Before proceeding, let’s remove genes that are not expressed in any of the cells. This helps in reducing the noise and computational load.
After filtering, we retain approximately 17,000 genes.
Creating the Design Matrix
Next, we create a design_matrix, which is a matrix that
includes the features of each cell. The model.matrix
function is an easy way to generate this matrix, requiring a formula and
the metadata as input.
design_matrix <- model.matrix(~Biological_Condition, data = metadata)
print(unique(metadata$Biological_Condition)) # Display unique biological conditions
#> [1] "GW16" "NPC" "GW21+3" "GW21"
head(design_matrix) # Display the first few rows of the design matrix
#> (Intercept) Biological_ConditionGW21 Biological_ConditionGW21+3
#> SRR1275356 1 0 0
#> SRR1274090 1 0 0
#> SRR1275251 1 0 0
#> SRR1275287 1 0 1
#> SRR1275364 1 0 0
#> SRR1275269 1 1 0
#> Biological_ConditionNPC
#> SRR1275356 0
#> SRR1274090 1
#> SRR1275251 0
#> SRR1275287 0
#> SRR1275364 0
#> SRR1275269 0Here, the design matrix models each cell based on its
Biological_Condition. The intercept in this model
represents the “GW16” condition.
Fitting the model
With the data prepared, we can now fit the model using the
fit_devil function. This function estimates the
coefficients (beta) and overdispersion for each gene across
different conditions.
fit <- devil::fit_devil(
as.matrix(counts),
design_matrix,
overdispersion = T,
size_factors = T,
verbose = T,
parallel.cores = 1
)
#> Compute size factors
#> Initialize beta estimate
#> Fit beta coefficients
#> Fit overdispersionThe fit object contains several important components,
including:
-
beta: matrix of coefficients (n_genes x n_features) -
overdispersion: vector of coefficients (n_genes)
Testing for Differential Expression
In order to test the data you need to specify your null hypothesis using a contrast vector . Considering a gene along with its inferred coefficient , the null hypothesis is usually defined as
For example, if you are interested in the genes that are differentially expressed between the “GW21” and the “NPC” condition, you need to find the genes for which we strongly reject the null hypothesis
$$ \beta_{GW21} = \beta_{NPC} \hspace{5mm}
\rightarrow \hspace{5mm} \beta_{GW21} - \beta_{NPC} = 0$$ which
is equivalent to defining the contrast vector
.
Once the contrast vector is defined, you can test the null hypothesis
using the test_de function.
contrast <- c(0, 1, 0, -1)
test_res <- devil::test_de(fit, contrast, max_lfc = 20)
if (!('name' %in% colnames(test_res))) {
test_res$name <- as.character( 1:nrow(test_res) )
}
head(test_res)
#> # A tibble: 6 × 4
#> pval adj_pval lfc name
#> <dbl> <dbl> <dbl> <chr>
#> 1 0.774 1 -18.8 1
#> 2 1.00 1 1.66 2
#> 3 0.997 1 1.66 3
#> 4 0.0115 0.137 6.70 4
#> 5 0.817 1 -19.4 5
#> 6 0.999 1 1.66 6The results contains, for each gene
-
pval: p-value associated with the statistical test -
adj_pval: p-value corrected considering multiple testing -
lfc: log2 fold change of gene expression between the tested conditions
Visualizing the Results with a Volcano Plot
Finally, we can visualize the results using a volcano plot, which
highlights genes based on their log-fold change (lfc) and
adjusted p-values (adj_pval). The plot_volcano
function from devil makes this easy.
devil::plot_volcano(test_res, lfc_cut = 1, pval_cut = .05, labels = TRUE, point_size = 2)
This plot provides a clear view of the most significant genes, making it easier to identify those that are likely to be biologically relevant.
Summary
In this vignette, we’ve demonstrated how to:
1. Load and preprocess scRNA-seq data.
2. Fit a model using the devil package.
3. Perform differential expression testing between conditions.
4. Visualize the results using a volcano plot.The devil package offers a powerful and flexible
framework for analyzing single-cell RNA sequencing data, helping
researchers uncover meaningful biological insights.