CiteFuse: getting started

Introduction

CiteFuse is a computational framework that implements a suite of methods and tools for CITE-seq data from pre-processing through to integrative analytics. This includes doublet detection, network-based modality integration, cell type clustering, differential RNA and protein expression analysis, ADT evaluation, ligand-receptor interaction analysis, and interactive web-based visualisation of the analyses. This vignette demonstrates the usage of CiteFuse on a subset data of CITE-seq data from human PBMCs as an example (Mimitou et al., 2019).

First, install CiteFuse using BiocManager.

if (!requireNamespace("BiocManager", quietly = TRUE)) {
 install.packages("BiocManager")
}
BiocManager::install("CiteFuse")
library(CiteFuse)
library(scater)
library(SingleCellExperiment)
library(DT)
data("CITEseq_example", package = "CiteFuse")
names(CITEseq_example)
#> [1] "RNA" "ADT" "HTO"
lapply(CITEseq_example, dim)
#> $RNA
#> [1] 19521   500
#> 
#> $ADT
#> [1]  49 500
#> 
#> $HTO
#> [1]   4 500

Here, we start from a list of three matrices of unique molecular identifier (UMI), antibody derived tags (ADT) and hashtag oligonucleotide (HTO) count, which have common cell names. There are 500 cells in our subsetted dataset. And characteristically of CITE-seq data, the matrices are matched, meaning that for any given cell we know the expression level of their RNA transcripts (genome-wide) and its corresponding cell surface protein expression. The preprocessing function will utilise the three matrices and its common cell names to create a SingleCellExperiment object, which stores RNA data in an assay and ADT and HTO data within in the altExp slot.

sce_citeseq <- preprocessing(CITEseq_example)
sce_citeseq
#> class: SingleCellExperiment 
#> dim: 19521 500 
#> metadata(0):
#> assays(1): counts
#> rownames(19521): hg19_AL627309.1 hg19_AL669831.5 ... hg19_MT-ND6
#>   hg19_MT-CYB
#> rowData names(0):
#> colnames(500): AAGCCGCGTTGTCTTT GATCGCGGTTATCGGT ... TTGGCAACACTAGTAC
#>   GCTGCGAGTTGTGGCC
#> colData names(0):
#> reducedDimNames(0):
#> mainExpName: NULL
#> altExpNames(2): ADT HTO

Detecting both cross- and within-sample doublets using CiteFuse

HTO Normalisation and Visualisation

The function normaliseExprs is used to scale the alternative expression. Here, we used it to perform log-transformation of the HTO count, by setting transform = "log".

sce_citeseq <- normaliseExprs(sce = sce_citeseq, 
                              altExp_name = "HTO", 
                              transform = "log")

Then we can perform dimension reduction on the HTO count by using runTSNE or runUMAP, then use visualiseDim function to visualise the reduced dimension plot. Our CITE-seq dataset contain data from four samples that were pooled before sequencing. The samples were multiplexed through cell hashing (Stoekius et al., 2018). The four clusters observed on reduced dimension plots equate to the four different samples.


sce_citeseq <- scater::runTSNE(sce_citeseq, 
                               altexp = "HTO", 
                               name = "TSNE_HTO", 
                               pca = TRUE)


visualiseDim(sce_citeseq,
             dimNames = "TSNE_HTO") + labs(title = "tSNE (HTO)")


sce_citeseq <- scater::runUMAP(sce_citeseq, 
                               altexp = "HTO", 
                               name = "UMAP_HTO")


visualiseDim(sce_citeseq,
             dimNames = "UMAP_HTO") + labs(title = "UMAP (HTO)")

Doublet identification step 1: cross-sample doublet detection

An important step in single cell data analysis is the removal of doublets. Doublets form as a result of co-encapsulation of cells within a droplet, leading to a hybrid transcriptome from two or more cells. In CiteFuse, we implement a step-wise doublet detection approach to remove doublets. We first identify the cross-sample doublets via the crossSampleDoublets function.

sce_citeseq <- crossSampleDoublets(sce_citeseq)
#> number of iterations= 20 
#> number of iterations= 24 
#> number of iterations= 46 
#> number of iterations= 50

The results of the cross sample doublets are then saved in colData as doubletClassify_between_label and doubletClassify_between_class.

table(sce_citeseq$doubletClassify_between_label)
#> 
#>                 1                 2                 3                 4 
#>               115               121                92               129 
#> doublet/multiplet 
#>                43
table(sce_citeseq$doubletClassify_between_class)
#> 
#>           Singlet doublet/multiplet 
#>               457                43

We can then highlight the cross-sample doublets in our tSNE plot of HTO count.

visualiseDim(sce_citeseq, 
             dimNames = "TSNE_HTO", 
             colour_by = "doubletClassify_between_label")

Furthermore, plotHTO function allows us to plot the pairwise scatter HTO count. Any cells that show co-expression of orthologocal HTOs (red) are considered as doublets.

plotHTO(sce_citeseq, 1:4)

Doublet identification step 1: within-sample doublet detection

We then identify the within-sample doublets via the withinSampleDoublets function.

sce_citeseq <- withinSampleDoublets(sce_citeseq,
                                    minPts = 10)

The results of the cross sample doublets are then saved in the colData as doubletClassify_within_label and doubletClassify_within_class.

table(sce_citeseq$doubletClassify_within_label)
#> 
#>  Doublets(Within)_1  Doublets(Within)_2  Doublets(Within)_3  Doublets(Within)_4 
#>                   3                   7                   4                   6 
#> NotDoublets(Within) 
#>                 480
table(sce_citeseq$doubletClassify_within_class)
#> 
#> Doublet Singlet 
#>      20     480

Again, we can visualise the within-sample doublets in our tSNE plot.

visualiseDim(sce_citeseq, 
             dimNames = "TSNE_HTO", 
             colour_by = "doubletClassify_within_label")

Finally, we can filter out the doublet cells (both within and between batches) for the downstream analysis.

sce_citeseq <- sce_citeseq[, sce_citeseq$doubletClassify_within_class == "Singlet" & sce_citeseq$doubletClassify_between_class == "Singlet"]
sce_citeseq
#> class: SingleCellExperiment 
#> dim: 19521 437 
#> metadata(3): doubletClassify_between_threshold
#>   doubletClassify_between_resultsMat doubletClassify_within_resultsMat
#> assays(1): counts
#> rownames(19521): hg19_AL627309.1 hg19_AL669831.5 ... hg19_MT-ND6
#>   hg19_MT-CYB
#> rowData names(0):
#> colnames(437): GATCGCGGTTATCGGT GGCTGGTAGAGGTTAT ... TTGGCAACACTAGTAC
#>   GCTGCGAGTTGTGGCC
#> colData names(5): doubletClassify_between_label
#>   doubletClassify_between_class nUMI doubletClassify_within_label
#>   doubletClassify_within_class
#> reducedDimNames(2): TSNE_HTO UMAP_HTO
#> mainExpName: NULL
#> altExpNames(2): ADT HTO

Clustering

Performing SNF

The first step of analysis is to integrate the RNA and ADT matrix. We use a popular integration algorithm called similarity network fusion (SNF) to integrate the multiomic data.

sce_citeseq <- scater::logNormCounts(sce_citeseq)
sce_citeseq <- normaliseExprs(sce_citeseq, altExp_name = "ADT", transform = "log")
system.time(sce_citeseq <- CiteFuse(sce_citeseq))
#> Calculating affinity matrix 
#> Performing SNF
#>    user  system elapsed 
#>   1.301   0.651   0.988

We now proceed with the fused matrix, which is stored as SNF_W in our sce_citeseq object.

Performing spectral clustering

CiteFuse implements two different clustering algorithms on the fused matrix, spectral clustering and Louvain clustering. First, we perform spectral clustering with sufficient numbers of K and use the eigen values to determine the optimal number of clusters.

SNF_W_clust <- spectralClustering(metadata(sce_citeseq)[["SNF_W"]], K = 20)
#> Computing Spectral Clustering
plot(SNF_W_clust$eigen_values)

which.max(abs(diff(SNF_W_clust$eigen_values)))
#> [1] 5

Using the optimal cluster number defined from the previous step, we can now use the spectralClutering function to cluster the single cells by specifying the number of clusters in K. The function takes a cell-to-cell similarity matrix as an input. We have already created the fused similarity matrix from CiteFuse. Since the CiteFuse function creates and stores the similarity matries from ADT and RNA expression, as well the fused matrix, we can use these two to compare the clustering outcomes by data modality.

SNF_W_clust <- spectralClustering(metadata(sce_citeseq)[["SNF_W"]], K = 5)
#> Computing Spectral Clustering
sce_citeseq$SNF_W_clust <- as.factor(SNF_W_clust$labels)

SNF_W1_clust <- spectralClustering(metadata(sce_citeseq)[["ADT_W"]], K = 5)
#> Computing Spectral Clustering
sce_citeseq$ADT_clust <- as.factor(SNF_W1_clust$labels)

SNF_W2_clust <- spectralClustering(metadata(sce_citeseq)[["RNA_W"]], K = 5)
#> Computing Spectral Clustering
sce_citeseq$RNA_clust <- as.factor(SNF_W2_clust$labels)

Visualisation

The outcome of the clustering can be easily visualised on a reduced dimensions plot by highlighting the points by cluster label.


sce_citeseq <- reducedDimSNF(sce_citeseq,
                             method = "tSNE", 
                             dimNames = "tSNE_joint")

g1 <- visualiseDim(sce_citeseq, dimNames = "tSNE_joint", colour_by = "SNF_W_clust") +
  labs(title = "tSNE (SNF clustering)")
g2 <- visualiseDim(sce_citeseq, dimNames = "tSNE_joint",  colour_by = "ADT_clust") +
  labs(title = "tSNE (ADT clustering)")
g3 <- visualiseDim(sce_citeseq, dimNames = "tSNE_joint",  colour_by = "RNA_clust") +
  labs(title = "tSNE (RNA clustering)")

library(gridExtra)
grid.arrange(g3, g2, g1, ncol = 2)

The expression of genes and proteins can be visualised by changing the colour_by parameter to assess the clusters. As an example, we highlight the plot by the RNA and ADT expression level of CD8.

g1 <- visualiseDim(sce_citeseq, dimNames = "tSNE_joint", 
                   colour_by = "hg19_CD8A",
                   data_from = "assay",
                   assay_name = "logcounts") +
  labs(title = "tSNE: hg19_CD8A (RNA expression)")
g2 <- visualiseDim(sce_citeseq,dimNames = "tSNE_joint", 
                   colour_by = "CD8",
                   data_from = "altExp",
                   altExp_assay_name = "logcounts") +
  labs(title = "tSNE: CD8 (ADT expression)")

grid.arrange(g1, g2, ncol = 2)

Louvain clustering

As well as spectral clustering, CiteFuse can implement Louvain clustering if users wish to use another clustering method. We use the igraph package, and any community detection algorithms available in their package can be selected by changing the method parameter.

SNF_W_louvain <- igraphClustering(sce_citeseq, method = "louvain")
table(SNF_W_louvain)
#> SNF_W_louvain
#>   1   2   3   4   5   6   7 
#>  88 139  62  32  51  29  36

sce_citeseq$SNF_W_louvain <- as.factor(SNF_W_louvain)

visualiseDim(sce_citeseq, dimNames = "tSNE_joint", colour_by = "SNF_W_louvain") +
  labs(title = "tSNE (SNF louvain clustering)")

visualiseKNN(sce_citeseq, colour_by = "SNF_W_louvain")

Differential Expression Analysis

Exploration of feature expression

CiteFuse has a wide range of visualisation tools to facilitate exploratory analysis of CITE-seq data. The visualiseExprs function is an easy-to-use function to generate boxplots, violinplots, jitter plots, density plots, and pairwise scatter/density plots of genes and proteins expressed in the data. The plots can be grouped by using the cluster labels stored in the sce_citeseq object.

visualiseExprs(sce_citeseq, 
               plot = "boxplot", 
               group_by = "SNF_W_louvain",
               feature_subset = c("hg19_CD2", "hg19_CD4", "hg19_CD8A", "hg19_CD19"))

visualiseExprs(sce_citeseq, 
               plot = "violin", 
               group_by = "SNF_W_louvain",
               feature_subset = c("hg19_CD2", "hg19_CD4", "hg19_CD8A", "hg19_CD19"))

visualiseExprs(sce_citeseq, 
               plot = "jitter", 
               group_by = "SNF_W_louvain",
               feature_subset = c("hg19_CD2", "hg19_CD4", "hg19_CD8A", "hg19_CD19"))

visualiseExprs(sce_citeseq, 
               plot = "density", 
               group_by = "SNF_W_louvain",
               feature_subset = c("hg19_CD2", "hg19_CD4", "hg19_CD8A", "hg19_CD19"))

visualiseExprs(sce_citeseq, 
               altExp_name = "ADT", 
               group_by = "SNF_W_louvain",
               plot = "violin", n = 5)

visualiseExprs(sce_citeseq, altExp_name = "ADT", 
               plot = "jitter", 
               group_by = "SNF_W_louvain", 
               feature_subset = c("CD2", "CD8", "CD4", "CD19"))

visualiseExprs(sce_citeseq, altExp_name = "ADT", 
               plot = "density", 
               group_by = "SNF_W_louvain",
               feature_subset = c("CD2", "CD8", "CD4", "CD19"))

visualiseExprs(sce_citeseq, altExp_name = "ADT", 
               plot = "pairwise", 
               feature_subset = c("CD4", "CD8"))
#> number of iterations= 25 
#> number of iterations= 24


visualiseExprs(sce_citeseq, altExp_name = "ADT", 
               plot = "pairwise", 
               feature_subset = c("CD45RA", "CD4", "CD8"), 
               threshold = rep(4, 3))

Perform DE Analysis with Wilcoxon Rank Sum test

CiteFuse also calculates differentially expressed (DE) genes through the DEgenes function. The cluster grouping to use must be specified in the group parameter. If altExp_name is not specified, RNA expression will be used as the default expression matrix.

Results form the DE analysis is stored in sce_citeseq as DE_res_RNA_filter and DE_res_ADT_filter for RNA and ADT expression, respectively.

For RNA expression

# DE will be performed for RNA if altExp_name = "none" 
sce_citeseq <- DEgenes(sce_citeseq,
                       altExp_name = "none", 
                       group = sce_citeseq$SNF_W_louvain,
                       return_all = TRUE,
                       exprs_pct = 0.5)


sce_citeseq <- selectDEgenes(sce_citeseq,
                             altExp_name = "none")
datatable(format(do.call(rbind, metadata(sce_citeseq)[["DE_res_RNA_filter"]]), 
                 digits = 2))

For ADT count

sce_citeseq <- DEgenes(sce_citeseq,
                       altExp_name = "ADT", 
                       group = sce_citeseq$SNF_W_louvain,
                       return_all = TRUE,
                       exprs_pct = 0.5)


sce_citeseq <- selectDEgenes(sce_citeseq,
                             altExp_name = "ADT")
datatable(format(do.call(rbind, metadata(sce_citeseq)[["DE_res_ADT_filter"]]), 
                 digits = 2))

Visualising DE Results

The DE genes can be visualised with the DEbubblePlot and DEcomparisonPlot. In each case, the gene names must first be extracted from the DE result objects.

circlepackPlot

The circlepackPlot takes a list of all DE genes from RNA and ADT DE analysis and will plot only the top most significant DE genes to plot.

rna_DEgenes <- metadata(sce_citeseq)[["DE_res_RNA_filter"]]
adt_DEgenes <- metadata(sce_citeseq)[["DE_res_ADT_filter"]]

rna_DEgenes <- lapply(rna_DEgenes, function(x){
  x$name <- gsub("hg19_", "", x$name)
  x})
DEbubblePlot(list(RNA = rna_DEgenes, ADT = adt_DEgenes))

DEcomparisonPlot

For the DEcomparisonPlot, as well as a list containing the DE genes for RNA and ADT, a feature_list specifying the genes and proteins of interest is required.

rna_list <- c("hg19_CD4",
              "hg19_CD8A",
              "hg19_HLA-DRB1",
              "hg19_ITGAX",
              "hg19_NCAM1",
              "hg19_CD27",
              "hg19_CD19")

adt_list <- c("CD4", "CD8", "MHCII (HLA-DR)", "CD11c", "CD56", "CD27", "CD19")

rna_DEgenes_all <- metadata(sce_citeseq)[["DE_res_RNA"]]
adt_DEgenes_all <- metadata(sce_citeseq)[["DE_res_ADT"]]

feature_list <- list(RNA = rna_list, ADT = adt_list)
de_list <- list(RNA = rna_DEgenes_all, ADT = adt_DEgenes_all)

DEcomparisonPlot(de_list = de_list,
                 feature_list = feature_list)

ADT Importance Evaluation

An important evaluation in CITE-seq data analysis is to assess the quality of each ADT and to evaluate the contribution of ADTs towards clustering outcome. CiteFuse calculates the relative importance of ADT towards clustering outcome by using a random forest model. The higher the score of an ADT, the greater its importance towards the final clustering outcome.

set.seed(2020)
sce_citeseq <- importanceADT(sce_citeseq, 
                             group = sce_citeseq$SNF_W_louvain,
                             subsample = TRUE)

visImportance(sce_citeseq, plot = "boxplot")

visImportance(sce_citeseq, plot = "heatmap")


sort(metadata(sce_citeseq)[["importanceADT"]], decreasing = TRUE)[1:20]
#>              CD27               CD8               CD4              CD28 
#>         39.470893         36.274249         33.235362         12.633975 
#>               CD5      PECAM (CD31)             CD11b               CD7 
#>         11.843918         11.780143         11.393725         11.304905 
#> IL7Ralpha (CD127)    MHCII (HLA-DR)               CD2              CD44 
#>         10.044518          8.732073          8.333783          7.781941 
#>      CD366 (tim3)         HLA-A,B,C             CD11c               CD3 
#>          6.437104          5.596078          5.238851          4.580886 
#>            CD45RA      PD-1 (CD279)              CD69       PD1 (CD279) 
#>          3.978908          3.834483          3.727649          3.570590

The importance scores can be visualised in a boxplot and heatmap. Our evaluation of ADT importance show that unsurprisingly CD4 and CD8 are the top two discriminating proteins in PBMCs.

Let us try clustering with only ADTs with a score greater than 5.

subset_adt <- names(which(metadata(sce_citeseq)[["importanceADT"]] > 5))
subset_adt
#>  [1] "CD11b"             "CD11c"             "CD2"              
#>  [4] "CD27"              "CD28"              "CD366 (tim3)"     
#>  [7] "CD4"               "CD44"              "CD5"              
#> [10] "CD7"               "CD8"               "HLA-A,B,C"        
#> [13] "IL7Ralpha (CD127)" "MHCII (HLA-DR)"    "PECAM (CD31)"

system.time(sce_citeseq <- CiteFuse(sce_citeseq,
                                    ADT_subset = subset_adt,
                                    metadata_names = c("W_SNF_adtSubset1",
                                                       "W_ADT_adtSubset1",
                                                       "W_RNA")))
#> Calculating affinity matrix 
#> Performing SNF
#>    user  system elapsed 
#>   1.203   0.411   0.910

SNF_W_clust_adtSubset1 <- spectralClustering(metadata(sce_citeseq)[["W_SNF_adtSubset1"]], K = 5)
#> Computing Spectral Clustering
sce_citeseq$SNF_W_clust_adtSubset1 <- as.factor(SNF_W_clust_adtSubset1$labels)


library(mclust)
adjustedRandIndex(sce_citeseq$SNF_W_clust_adtSubset1, sce_citeseq$SNF_W_clust)
#> [1] 0.8646855

When we compare between the two clustering outcomes, we find that the adjusted rand index is approximately 0.93, where a value of 1 denotes complete concordance.

Gene - ADT network

The geneADTnetwork function plots an interaction network between genes identified from the DE analysis. The nodes denote proteins and RNA whilst the edges denote positive and negative correlation in expression.

RNA_feature_subset <- unique(as.character(unlist(lapply(rna_DEgenes_all, "[[", "name"))))
ADT_feature_subset <- unique(as.character(unlist(lapply(adt_DEgenes_all, "[[", "name"))))

geneADTnetwork(sce_citeseq,
               RNA_feature_subset = RNA_feature_subset,
               ADT_feature_subset = ADT_feature_subset,
               cor_method = "pearson",
               network_layout = igraph::layout_with_fr)

#> IGRAPH e91d228 UN-B 72 134 -- 
#> + attr: name (v/c), label (v/c), class (v/c), type (v/l), shape (v/c),
#> | color (v/c), size (v/n), label.cex (v/n), label.color (v/c), value
#> | (e/n), color (e/c), weights (e/n)
#> + edges from e91d228 (vertex names):
#>  [1] RNA_hg19_CD8A --ADT_CD4  RNA_hg19_CCL5 --ADT_CD4  RNA_hg19_GNLY --ADT_CD4 
#>  [4] RNA_hg19_KLRD1--ADT_CD4  RNA_hg19_GZMB --ADT_CD4  RNA_hg19_CST7 --ADT_CD4 
#>  [7] RNA_hg19_NKG7 --ADT_CD4  RNA_hg19_CTSW --ADT_CD4  RNA_hg19_LTB  --ADT_CD27
#> [10] RNA_hg19_TCF7 --ADT_CD27 RNA_hg19_RPL32--ADT_CD27 RNA_hg19_IL7R --ADT_CD27
#> [13] RNA_hg19_RPL13--ADT_CD27 RNA_hg19_RPL37--ADT_CD27 RNA_hg19_RPS8 --ADT_CD27
#> [16] RNA_hg19_RPL11--ADT_CD27 RNA_hg19_CD27 --ADT_CD27 RNA_hg19_RPS12--ADT_CD27
#> + ... omitted several edges

RNA Ligand - ADT Receptor Analysis

With the advent of CITE-seq, we can now predict ligand-receptor interactions by using cell surface protein expression. CiteFuse implements a ligandReceptorTest to find ligand receptor interactions between sender and receiver cells. Importantly, the ADT count is used to predict receptor expression within receiver cells. Note that the setting altExp_name = "RNA" would enable users to predict ligand-receptor interaction from RNA expression only.

data("lr_pair_subset", package = "CiteFuse")
head(lr_pair_subset)
#>      [,1]          [,2]   
#> [1,] "hg19_IL17RA" "CD45" 
#> [2,] "hg19_FAS"    "CD11b"
#> [3,] "hg19_GZMK"   "CD62L"
#> [4,] "hg19_CD40LG" "CD11b"
#> [5,] "hg19_FLT3LG" "CD62L"
#> [6,] "hg19_GZMA"   "CD19"

sce_citeseq <- normaliseExprs(sce = sce_citeseq, 
                              altExp_name = "ADT", 
                              transform = "zi_minMax")

sce_citeseq <- normaliseExprs(sce = sce_citeseq, 
                              altExp_name = "none", 
                              exprs_value = "logcounts",
                              transform = "minMax")

sce_citeseq <- ligandReceptorTest(sce = sce_citeseq,
                                  ligandReceptor_list = lr_pair_subset,
                                  cluster = sce_citeseq$SNF_W_louvain,
                                  RNA_exprs_value = "minMax",
                                  use_alt_exp = TRUE,
                                  altExp_name = "ADT",
                                  altExp_exprs_value = "zi_minMax",
                                  num_permute = 1000) 
#> 100 ......200 ......300 ......400 ......500 ......600 ......700 ......800 ......900 ......1000 ......
visLigandReceptor(sce_citeseq, 
                  type = "pval_heatmap",
                  receptor_type = "ADT")

visLigandReceptor(sce_citeseq, 
                  type = "pval_dotplot",
                  receptor_type = "ADT")

visLigandReceptor(sce_citeseq, 
                  type = "group_network",
                  receptor_type = "ADT")

visLigandReceptor(sce_citeseq, 
                  type = "group_heatmap",
                  receptor_type = "ADT")

visLigandReceptor(sce_citeseq,
                  type = "lr_network",
                  receptor_type = "ADT")

Between-sample analysis

Lastly, we will jointly analyse the current PBMC CITE-seq data, taken from healthy human donors, and another subset of CITE-seq data from patients with cutaneous T-cell lymphoma (CTCL), again from Mimitou et al. (2019). The data sce_ctcl_subset provided in our CiteFuse package already contains the clustering information.

data("sce_ctcl_subset", package = "CiteFuse")

To visualise and compare gene or protein expression data, we can use visualiseExprsList function.


visualiseExprsList(sce_list = list(control = sce_citeseq,
                                   ctcl = sce_ctcl_subset),
                   plot = "boxplot",
                   altExp_name = "none",
                   exprs_value = "logcounts",
                   feature_subset = c("hg19_S100A10", "hg19_CD8A"),
                   group_by = c("SNF_W_louvain", "SNF_W_louvain"))


visualiseExprsList(sce_list = list(control = sce_citeseq,
                                   ctcl = sce_ctcl_subset),
                   plot = "boxplot",
                   altExp_name = "ADT", 
                   feature_subset = c("CD19", "CD8"),
                   group_by = c("SNF_W_louvain", "SNF_W_louvain"))

We can then perform differential expression analysis of the RNA expression level across the two clusters that have high CD19 expression in ADT.

de_res <- DEgenesCross(sce_list = list(control = sce_citeseq,
                                       ctcl = sce_ctcl_subset),
                       colData_name = c("SNF_W_louvain", "SNF_W_louvain"),
                       group_to_test = c("2", "6"))
de_res_filter <- selectDEgenes(de_res = de_res)
de_res_filter
#> $control
#>             stats.W         pval     p.adjust meanExprs.1 meanExprs.2
#> hg19_GNLY      28.0 5.642738e-23 5.642738e-23  0.07220401    4.585384
#> hg19_CST7     152.5 3.913672e-21 3.913672e-21  0.22512314    2.630719
#> hg19_NKG7     168.0 7.860013e-21 7.860013e-21  0.51066894    3.812262
#> hg19_GZMH     301.0 1.567852e-19 1.567852e-19  0.00000000    1.967411
#> hg19_GZMB     498.0 5.422652e-17 5.422652e-17  0.11667246    2.006243
#> hg19_CCL5     477.0 8.039376e-17 8.039376e-17  1.04348029    3.777290
#> hg19_FGFBP2   571.0 2.154067e-16 2.154067e-16  0.03101677    1.651125
#> hg19_KLRD1    598.0 6.180246e-16 6.180246e-16  0.09327555    1.564246
#> hg19_EFHD2    794.0 3.699717e-14 3.699717e-14  0.02064028    1.210523
#> hg19_PRF1     803.5 6.796491e-14 6.796491e-14  0.06473009    1.461993
#>              meanPct.1 meanPct.2 meanDiff   pctDiff        name   group
#> hg19_GNLY   0.02325581 0.9928058 4.513180 0.9695499   hg19_GNLY control
#> hg19_CST7   0.13953488 0.9928058 2.405596 0.8532709   hg19_CST7 control
#> hg19_NKG7   0.30232558 1.0000000 3.301593 0.6976744   hg19_NKG7 control
#> hg19_GZMH   0.00000000 0.8992806 1.967411 0.8992806   hg19_GZMH control
#> hg19_GZMB   0.06976744 0.8776978 1.889570 0.8079304   hg19_GZMB control
#> hg19_CCL5   0.37209302 1.0000000 2.733809 0.6279070   hg19_CCL5 control
#> hg19_FGFBP2 0.02325581 0.8201439 1.620108 0.7968881 hg19_FGFBP2 control
#> hg19_KLRD1  0.09302326 0.8273381 1.470971 0.7343149  hg19_KLRD1 control
#> hg19_EFHD2  0.02325581 0.7410072 1.189883 0.7177514  hg19_EFHD2 control
#> hg19_PRF1   0.04651163 0.7625899 1.397263 0.7160783   hg19_PRF1 control
#> 
#> $ctcl
#>               stats.W         pval     p.adjust meanExprs.1 meanExprs.2
#> hg19_LTB        372.0 3.139096e-28 3.139096e-28  0.10343246    1.978730
#> hg19_RPS26        0.0 4.172127e-23 4.172127e-23  1.64098856    5.072471
#> hg19_IL7R       780.0 1.412936e-20 1.412936e-20  0.17329085    1.651591
#> hg19_SELL      1033.5 8.045007e-20 8.045007e-20  0.07322948    1.152572
#> hg19_LEPROTL1   715.0 2.219774e-18 2.219774e-18  0.24294829    1.300745
#> hg19_EEF1B2     523.5 3.123434e-16 3.123434e-16  1.71942732    3.043623
#> hg19_HBB       1737.5 1.285550e-15 1.285550e-15  0.00000000    0.431026
#> hg19_NOSIP     1044.0 1.332385e-14 1.332385e-14  0.19852399    1.157235
#> hg19_FOS       1324.0 3.386001e-13 3.386001e-13  0.16536118    1.125337
#> hg19_NPM1       945.0 1.042899e-11 1.042899e-11  1.17875475    2.202268
#>                meanPct.1 meanPct.2  meanDiff   pctDiff          name group
#> hg19_LTB      0.07913669 0.9069767 1.8752977 0.8278401      hg19_LTB  ctcl
#> hg19_RPS26    0.89208633 1.0000000 3.4314820 0.1079137    hg19_RPS26  ctcl
#> hg19_IL7R     0.10791367 0.8139535 1.4782997 0.7060398     hg19_IL7R  ctcl
#> hg19_SELL     0.05755396 0.6976744 1.0793420 0.6401205     hg19_SELL  ctcl
#> hg19_LEPROTL1 0.19424460 0.9069767 1.0577972 0.7127321 hg19_LEPROTL1  ctcl
#> hg19_EEF1B2   0.86330935 0.9767442 1.3241960 0.1134348   hg19_EEF1B2  ctcl
#> hg19_HBB      0.00000000 0.4186047 0.4310260 0.4186047      hg19_HBB  ctcl
#> hg19_NOSIP    0.19424460 0.7674419 0.9587114 0.5731973    hg19_NOSIP  ctcl
#> hg19_FOS      0.12230216 0.6511628 0.9599762 0.5288606      hg19_FOS  ctcl
#> hg19_NPM1     0.73381295 0.9534884 1.0235132 0.2196754     hg19_NPM1  ctcl

Read data from 10X Genomics

Readers unfamiliar with the workflow of converting a count matrix into a SingleCellExperiment object may use the readFrom10X function to convert count matrix from a 10X experiment into an object that can be used for all functions in CiteFuse.

tmpdir <- tempdir()
download.file("http://cf.10xgenomics.com/samples/cell-exp/3.1.0/connect_5k_pbmc_NGSC3_ch1/connect_5k_pbmc_NGSC3_ch1_filtered_feature_bc_matrix.tar.gz", file.path(tmpdir, "/5k_pbmc_NGSC3_ch1_filtered_feature_bc_matrix.tar.gz"))
untar(file.path(tmpdir, "5k_pbmc_NGSC3_ch1_filtered_feature_bc_matrix.tar.gz"),
      exdir = tmpdir)
sce_citeseq_10X <- readFrom10X(file.path(tmpdir, "filtered_feature_bc_matrix/"))
sce_citeseq_10X

SessionInfo

sessionInfo()
#> R version 4.5.2 (2025-10-31)
#> Platform: x86_64-pc-linux-gnu
#> Running under: Ubuntu 24.04.3 LTS
#> 
#> Matrix products: default
#> BLAS:   /usr/lib/x86_64-linux-gnu/openblas-pthread/libblas.so.3 
#> LAPACK: /usr/lib/x86_64-linux-gnu/openblas-pthread/libopenblasp-r0.3.26.so;  LAPACK version 3.12.0
#> 
#> locale:
#>  [1] LC_CTYPE=en_US.UTF-8       LC_NUMERIC=C              
#>  [3] LC_TIME=en_US.UTF-8        LC_COLLATE=C              
#>  [5] LC_MONETARY=en_US.UTF-8    LC_MESSAGES=en_US.UTF-8   
#>  [7] LC_PAPER=en_US.UTF-8       LC_NAME=C                 
#>  [9] LC_ADDRESS=C               LC_TELEPHONE=C            
#> [11] LC_MEASUREMENT=en_US.UTF-8 LC_IDENTIFICATION=C       
#> 
#> time zone: Etc/UTC
#> tzcode source: system (glibc)
#> 
#> attached base packages:
#> [1] stats4    stats     graphics  grDevices utils     datasets  methods  
#> [8] base     
#> 
#> other attached packages:
#>  [1] mclust_6.1.2                gridExtra_2.3              
#>  [3] DT_0.34.0                   scater_1.39.2              
#>  [5] ggplot2_4.0.1               scuttle_1.21.0             
#>  [7] SingleCellExperiment_1.33.0 SummarizedExperiment_1.41.0
#>  [9] Biobase_2.71.0              GenomicRanges_1.63.1       
#> [11] Seqinfo_1.1.0               IRanges_2.45.0             
#> [13] S4Vectors_0.49.0            BiocGenerics_0.57.0        
#> [15] generics_0.1.4              MatrixGenerics_1.23.0      
#> [17] matrixStats_1.5.0           CiteFuse_1.23.0            
#> [19] BiocStyle_2.39.0           
#> 
#> loaded via a namespace (and not attached):
#>   [1] rlang_1.1.7          magrittr_2.0.4       otel_0.2.0          
#>   [4] ggridges_0.5.7       compiler_4.5.2       vctrs_0.7.1         
#>   [7] reshape2_1.4.5       stringr_1.6.0        pkgconfig_2.0.3     
#>  [10] fastmap_1.2.0        XVector_0.51.0       labeling_0.4.3      
#>  [13] ggraph_2.2.2         rmarkdown_2.30       ggbeeswarm_0.7.3    
#>  [16] purrr_1.2.1          bluster_1.21.0       xfun_0.56           
#>  [19] beachmat_2.27.2      randomForest_4.7-1.2 cachem_1.1.0        
#>  [22] jsonlite_2.0.0       rhdf5filters_1.23.3  DelayedArray_0.37.0 
#>  [25] BiocParallel_1.45.0  Rhdf5lib_1.33.0      tweenr_2.0.3        
#>  [28] cluster_2.1.8.1      irlba_2.3.5.1        parallel_4.5.2      
#>  [31] R6_2.6.1             bslib_0.10.0         stringi_1.8.7       
#>  [34] RColorBrewer_1.1-3   limma_3.67.0         compositions_2.0-9  
#>  [37] jquerylib_0.1.4      Rcpp_1.1.1           knitr_1.51          
#>  [40] mixtools_2.0.0.1     FNN_1.1.4.1          Matrix_1.7-4        
#>  [43] splines_4.5.2        igraph_2.2.1         tidyselect_1.2.1    
#>  [46] abind_1.4-8          yaml_2.3.12          viridis_0.6.5       
#>  [49] codetools_0.2-20     lattice_0.22-7       tibble_3.3.1        
#>  [52] plyr_1.8.9           withr_3.0.2          S7_0.2.1            
#>  [55] evaluate_1.0.5       Rtsne_0.17           survival_3.8-6      
#>  [58] bayesm_3.1-7         polyclip_1.10-7      kernlab_0.9-33      
#>  [61] pillar_1.11.1        BiocManager_1.30.27  tensorA_0.36.2.1    
#>  [64] plotly_4.12.0        dbscan_1.2.4         scales_1.4.0        
#>  [67] glue_1.8.0           metapod_1.19.1       pheatmap_1.0.13     
#>  [70] lazyeval_0.2.2       maketools_1.3.2      tools_4.5.2         
#>  [73] BiocNeighbors_2.5.2  robustbase_0.99-6    sys_3.4.3           
#>  [76] data.table_1.18.2.1  RSpectra_0.16-2      ScaledMatrix_1.19.0 
#>  [79] locfit_1.5-9.12      scran_1.39.0         buildtools_1.0.0    
#>  [82] graphlayouts_1.2.2   tidygraph_1.3.1      cowplot_1.2.0       
#>  [85] rhdf5_2.55.12        grid_4.5.2           tidyr_1.3.2         
#>  [88] crosstalk_1.2.2      edgeR_4.9.2          nlme_3.1-168        
#>  [91] beeswarm_0.4.0       BiocSingular_1.27.1  ggforce_0.5.0       
#>  [94] vipor_0.4.7          rsvd_1.0.5           cli_3.6.5           
#>  [97] segmented_2.2-1      S4Arrays_1.11.1      viridisLite_0.4.2   
#> [100] dplyr_1.1.4          uwot_0.2.4           gtable_0.3.6        
#> [103] DEoptimR_1.1-4       sass_0.4.10          digest_0.6.39       
#> [106] dqrng_0.4.1          SparseArray_1.11.10  ggrepel_0.9.6       
#> [109] htmlwidgets_1.6.4    farver_2.1.2         memoise_2.0.1       
#> [112] htmltools_0.5.9      lifecycle_1.0.5      httr_1.4.7          
#> [115] statmod_1.5.1        MASS_7.3-65