Rcommands.Rmd

---
title: "scATAC-seq data analysis workflow"
output: html_document
date: "2023-08-21"
---

```{r setup, include=FALSE}
knitr::opts_chunk$set(echo = TRUE)
```

## Processed dataset
The final Seurat object resulting from the processing below can also be obtained from figshare (doi: 10.6084/m9.figshare.24040575).

After saving the file on your PC, this object can be loaded as follows:

```{r readRDS}
pbmc <- readRDS("processed.rds")
```

If you use this object, you can skip many steps below (including pre-processing, normalization, linear and non-linear dimensionality reduction, clustering).


## Load libraries
We need Seurat, Signac and EnsDb.Mmusculus.v75 for the single-cell ATAC-seq data analysis. 
Regarding Signac, refer to these URLs:
- https://stuartlab.org/signac/
- https://stuartlab.org/signac/articles/install.html
- https://stuartlab.org/signac/articles/overview.html
- https://stuartlab.org/signac/articles/pbmc_vignette.html
- https://stuartlab.org/signac/articles/mouse_brain_vignette
- https://stuartlab.org/signac/articles/motif_vignette

We need JASPAR2020, TFBSTools, and BSgenome.Mmusculus.UCSC.mm10 for motif analysis.
We need ggplot2 for making plots.


```{r libraries}
library(Signac)
library(Seurat)
library(EnsDb.Mmusculus.v75)

library(JASPAR2020)
library(TFBSTools)
library(BSgenome.Mmusculus.UCSC.mm10)

library(ggplot2)
library(reshape2)
```

## Set color scale
Set the color scale for plots (although this could be done using the Spectral palette as well...) and the colors of the cell clusters.

```{r color}
col.scale <- c('#5E4FA2', '#3F96B7', '#88CFA4', '#D7EF9B', '#FFFFBF', '#FDD380', '#F88D51', '#DC494C', '#9E0142')
cell.type.colors <-cols <- c(
  "#696969",     # C1, 
  "grey",        # C2
  "#414141",     # C3
  "darkgrey",    # C4
  "#e79478",     # C5
  "#d4582f",     # C6
  "#cc5b5b",     # C7
  "darkorange",  # C8
  "darkred",     # C9
  "#22bd1c",     # C10
  "#77bf75",     # C11
  "#6bf765",     # C12
  "#377d34",     # C13
  "#51ba4d",     # C14
  "#9c9cff",     # C15
  "#5f5f96",     # C16
  "#8282d1",     # C17
  "grey90"       # C18 (minor clusters)
)

```


## Pre-processing workflow

```{r pre}
# Read in the output of the 10X Genomics cellranger-atac software. 
DATA_IN <- "outs/" # the dir containing the cellranger-atac output
infile <- paste0(DATA_IN,"filtered_peak_bc_matrix.h5")
counts <- Read10X_h5(filename = infile)
dim(counts)

# Remove "strange" chromosomes.
counts.regions <- GRanges(rownames(counts))
length(counts.regions)
counts.regions.filtered <- keepStandardChromosomes(counts.regions, pruning.mode = "coarse")
length(counts.regions.filtered)
counts <- counts[as.character(counts.regions.filtered),]
dim(counts)


# Prepare meta data
infile <- paste0(DATA_IN,"singlecell.csv")
metadata <- read.csv(
  file = infile,
  header = TRUE,
  row.names = 1
)

metadata$sample <- NA
metadata$sample[grep(pattern="-1", x=rownames(metadata), value=FALSE)] <- "control"
metadata$sample[grep(pattern="-2", x=rownames(metadata), value=FALSE)] <- "Nfkbiz_KO"
metadata$sample[grep(pattern="-3", x=rownames(metadata), value=FALSE)] <- "DKO"
metadata$sample[grep(pattern="-4", x=rownames(metadata), value=FALSE)] <- "TKO"
metadata$sample <- factor(metadata$sample, 
                          levels = c("control", "DKO","Nfkbiz_KO", "TKO"))

# Read in fragment data
infile <- paste0(DATA_IN,"fragments.tsv.gz")
chrom_assay <- CreateChromatinAssay(
  counts = counts,
  sep = c(":", "-"),
  genome = 'mm10',
  fragments = infile,
  min.cells = 10,
  min.features = 200
)

# Prepare Seurat objects
# although we are not looking at PBMCs, let's just keep calling this variable pbmc...
pbmc <- CreateSeuratObject(
  counts = chrom_assay,
  assay = "peaks",
  meta.data = metadata
)
pbmc
```

The above results in a Seurat object of about 42k cells and 195k genomic regions.

Clean up some memory.
```{r cleanup}
rm(counts, chrom_assay, metadata, counts.regions, counts.regions.filtered)
gc()
```


Next, add gene annotations to the Seurat object using the mouse genome. 
```{r annotations}
# extract gene annotations from EnsDb
annotations <- GetGRangesFromEnsDb(ensdb = EnsDb.Mmusculus.v75)

# change to UCSC style since the data was mapped to mm10
# seqlevelsStyle(annotations) <- 'UCSC'
# However, lately this gives the following error:
# Error in function (type, msg, asError = TRUE)  : 
#   Failed to connect to ftp.ncbi.nlm.nih.gov port 21 after 42248 ms: Timed out
# So, instead:
seqlevels(annotations) <- paste0('chr', seqlevels(annotations))
genome(annotations) <- "mm10"

# add the gene information to the object
Annotation(pbmc) <- annotations
```

Might be good to clear up some memory.
```{r cleanup}
rm(annotations)
gc()
```

## Compute QC Metrics

```{r QC}
# compute nucleosome signal score per cell
pbmc <- NucleosomeSignal(object = pbmc)

# compute TSS enrichment score per cell
pbmc <- TSSEnrichment(object = pbmc, fast = FALSE)

# add blacklist ratio and fraction of reads in peaks
pbmc$pct_reads_in_peaks <- pbmc$peak_region_fragments / pbmc$passed_filters * 100
pbmc$blacklist_ratio <- pbmc$blacklist_region_fragments / pbmc$peak_region_fragments

pbmc$high.tss <- ifelse(pbmc$TSS.enrichment > 2, 'High', 'Low')
TSSPlot(pbmc, group.by = 'high.tss') + NoLegend()

pbmc$nucleosome_group <- ifelse(pbmc$nucleosome_signal > 4, 'NS > 4', 'NS < 4')
FragmentHistogram(object = pbmc, group.by = 'nucleosome_group')

VlnPlot(
  object = pbmc,
  features = c('pct_reads_in_peaks', 'peak_region_fragments',
               'TSS.enrichment', 'blacklist_ratio', 'nucleosome_signal'),
  pt.size = 0.1,
  ncol = 5
)

VlnPlot(
  object = pbmc,
  features = c('pct_reads_in_peaks', 'peak_region_fragments',
               'TSS.enrichment', 'blacklist_ratio', 'nucleosome_signal'),
  pt.size = 0,
  ncol = 5
)

VlnPlot(
  object = pbmc,
  features = c('pct_reads_in_peaks', 'peak_region_fragments',
               'TSS.enrichment', 'blacklist_ratio', 'nucleosome_signal'),
  pt.size = 0,
  ncol = 5,
  group.by = "sample"
)


# finally we remove cells that are outliers for these QC metrics.
pbmc <- subset(
  x = pbmc,
  subset = peak_region_fragments > 3000 &
    peak_region_fragments < 20000 &
    pct_reads_in_peaks > 30 &          # 15 in the example vignette
    nucleosome_signal < 4 &
    TSS.enrichment > 2
)
pbmc
```


## Normalization and linear dimensional reduction
```{r linear}
pbmc <- RunTFIDF(pbmc)
pbmc <- FindTopFeatures(pbmc, min.cutoff = 'q0')
pbmc <- RunSVD(pbmc)

DepthCor(pbmc)
# here too, as in the example vignettes: strong correlation between the first LSI and total number of counts per cell
```

## Non-linear dimension reduction and clustering
Note: excluding the first dimension (see comment at the end of the chunk above).

```{r nonlinear}
pbmc <- RunUMAP(object = pbmc, reduction = 'lsi', dims = 2:30)
pbmc <- FindNeighbors(object = pbmc, reduction = 'lsi', dims = 2:30)
pbmc <- FindClusters(object = pbmc, verbose = FALSE, algorithm = 3)
```


## Cluster annotation
```{r annotation}
manual_clusters <- rep(NA, length(pbmc$seurat_clusters))
manual_clusters[pbmc$seurat_clusters ==  0] <- "C1"
manual_clusters[pbmc$seurat_clusters ==  1] <- "C1"
manual_clusters[pbmc$seurat_clusters ==  3] <- "C2"
manual_clusters[pbmc$seurat_clusters == 12] <- "C3"
manual_clusters[pbmc$seurat_clusters == 18] <- "C4"
manual_clusters[pbmc$seurat_clusters ==  2] <- "C5"
# erythrocyte-oriented progenitors
manual_clusters[pbmc$seurat_clusters ==  9] <- "C6"
manual_clusters[pbmc$seurat_clusters ==  6] <- "C7"
manual_clusters[pbmc$seurat_clusters ==  4] <- "C8"
manual_clusters[pbmc$seurat_clusters ==  7] <- "C8"
manual_clusters[pbmc$seurat_clusters == 17] <- "C9"
# myeloid-oriented progenitors
manual_clusters[pbmc$seurat_clusters ==  8] <- "C10"
manual_clusters[pbmc$seurat_clusters == 13] <- "C11"
manual_clusters[pbmc$seurat_clusters == 10] <- "C12"
manual_clusters[pbmc$seurat_clusters ==  5] <- "C13"
manual_clusters[pbmc$seurat_clusters == 11] <- "C14"
# remaining minor clusters
manual_clusters[pbmc$seurat_clusters == 14] <- "C15"
manual_clusters[pbmc$seurat_clusters == 15] <- "C16"
manual_clusters[pbmc$seurat_clusters == 16] <- "C17"
manual_clusters[pbmc$seurat_clusters == 19] <- "C18"
manual_clusters[pbmc$seurat_clusters == 20] <- "C18"
manual_clusters[pbmc$seurat_clusters == 21] <- "C18"


pbmc$manual_clusters <- factor(manual_clusters, levels = c(
  "C1",  "C2",  "C3",  "C4" , "C5",
  "C6",  "C7",  "C8",  "C9" , 
  "C10", "C11", "C12", "C13", "C14",
  "C15", "C16", "C17", "C18"
))
table(pbmc$manual_clusters)

# some plots for Figures 5B,D
DimPlot(object = pbmc, group.by = "manual_clusters", label = FALSE, cols = cols)
DimPlot(object = pbmc, group.by = "manual_clusters", label = TRUE, cols = cols) + NoLegend()
DimPlot(object = pbmc, label = FALSE, split.by = "sample", group.by = "manual_clusters", cols = cols) + NoLegend()
```

## Create a gene activity matrix
```{r genes}
gene.activities <- GeneActivity(pbmc)
# add the gene activity matrix to the Seurat object as a new assay and normalize it
pbmc[['RNA']] <- CreateAssayObject(counts = gene.activities)
pbmc <- NormalizeData(
  object = pbmc,
  assay = 'RNA',
  normalization.method = 'LogNormalize',
  scale.factor = median(pbmc$nCount_RNA)
)
```

## Adding motif information to the Seurat object
```{r motifs}
# Get a list of motif position frequency matrices from the JASPAR database
pfm <- getMatrixSet(
  x = JASPAR2020,
  opts = list(collection = "CORE", tax_group = 'vertebrates', all_versions = FALSE)
)

# add motif information
DefaultAssay(pbmc) <- 'peaks' # if you don't do this, it will use 'RNA' leading to errors
pbmc <- AddMotifs(
  object = pbmc,
  genome = BSgenome.Mmusculus.UCSC.mm10,
  pfm = pfm
)

```


## Save what we have at the moment
```{r save}
processed.file <- paste0("processed.rds")
saveRDS(object = pbmc, file = processed.file)
```
The above processed.rds file can be found on figshare (doi: 10.6084/m9.figshare.24040575).


## UMAP of selected genes
Figures shown in Supplement for HSC, Myeloid, Meg/erythroid, and Lymphoid markers:
```{r selection}
genes <- c("Hlf",   "Meis1", "Mllt3",  "Pbx1",
           "Spi1",  "Csf1r", "S100a8", "F13a1",
           "Gata1", "Car1",  "Epor",   "Pf4",
           "Satb1", "Flt3",  "Dntt",   "Id4")
for(gene in genes){
  p <- FeaturePlot(
    object = pbmc,
    features = gene,
    pt.size = 0.1,
    max.cutoff = 'q95',
  ) + scale_colour_gradientn(colours = col.scale)
  ggsave(plot=p, file="feature_plot_",gene,".pdf", width=6, height=6)
}
```


## bar plot of cluster composition of each sample
The barplot shown in Figure 5C:

```{r barplot_samples}
tab <- as.matrix(table(pbmc$manual_clusters, pbmc$sample))
tab.col.tot <- apply(tab,2,sum)
tab.freq <- tab
for(c in 1:ncol(tab)){
  tab.freq[,c] <- tab[,c] / tab.col.tot[c] * 100
}
# confirm that this sums to 100
apply(tab.freq,2,sum)

tab.long <- melt(tab.freq, varnames = c("cluster", "sample"))
head(tab.long)

ggplot(tab.long, aes(x=sample,y=value,fill=cluster)) +
  geom_bar(position="stack", stat="identity") + theme_bw() +
  scale_fill_manual("legend", values = cols) +
  theme(axis.text.x = element_text(angle = 90, vjust = 0.5, hjust=1)) +
  theme(panel.grid.major.x = element_blank(), panel.grid.minor.x = element_blank())

```


## bar plot of sample composition of each cluster
The barplot at the bottom of Figure 5E:
```{r barplot}
tab <- as.matrix(table(pbmc$sample, pbmc$manual_clusters))
tab.row.tot <- apply(tab,1,sum)
tab.norm <- tab / tab.row.tot
tab.col.tot <- apply(tab.norm,2,sum)
tab.freq <- tab.norm
for(c in 1:ncol(tab)){
  tab.freq[,c] <- tab.norm[,c] / tab.col.tot[c] * 100
}

tab.long <- melt(tab.freq, varnames = c("sample", "cluster"))
head(tab.long)
tab.long$cluster <- factor(tab.long$cluster, levels=levels(pbmc$manual_clusters))
# to exclude the smaller clusters
tab.long <- subset(tab.long, !is.element(tab.long$cluster, c("C15","C16", "C17", "C18") ))
ggplot(tab.long, aes(x=cluster,y=value,fill=sample)) +
  geom_bar(position="stack", stat="identity") + theme_bw() +
  scale_fill_manual("legend", values = c("control"   = "blue", 
                                         "DKO"       = "red", 
                                         "Nfkbiz_KO" = "#4eac5b", # green
                                         "TKO"       = "#ff00ff") # pink
                    ) +
  theme(axis.text.x = element_text(angle = 90, vjust = 0.5, hjust=1))
```

## Heatmap of marker genes
The heatmap of Figure 5E:

```{r heatmapE}
markers <- c("Gprc5c", "Egr1",   "Neo1",   "Igf2bp2", "Ptges",
             "Mllt3",  "Msi2",   "Gata2",  "Hoxa9",   "Hoxa10",
             "Hoxb5",  "Meis1",  "Itga2b", "Hlf",     "Ifitm1",
             "Gstm1",  "Pbx1",   "Mn1",    "Lmo4",    "Cebpa",
             "Cebpb",  "Cebpe",  "Spi1",   "Csf1r",   "Csf2ra",
             "Csf2rb", "Csf3r",  "Fcer1g", "Cd52",    "F7",
             "F13a1",  "Ly86",   "Mpo",    "Elane",   "Fcnb",
             "Ltf",    "S100a8", "S100a9", "Cd74",    "Satb1",
             "Notch1", "Flt3",   "Dntt",   "Id4",     "Cd28",
             "Pf4",    "Gata1",   "Epor",  "Car1")
# get average expression of markers per cluster
averages <- AverageExpression(pbmc, assays = "RNA", group.by = "manual_clusters", features = markers)
dim(averages$RNA)
averages <- averages$RNA
# exclude the minor clusters
averages <- averages[, !is.element(colnames(averages), c("C15","C16","C17","C18"))]
averages[1:5,]

# scale
averages.norm <- t(scale(t(averages)))
# melt to long form
dat.melt <- melt(averages.norm, varnames = c("feature", "cluster"))
# to reverse the order in the heatmap
dat.melt$feature <- factor(dat.melt$feature, 
                             levels = rev(rownames(averages.norm)))
ggplot(dat.melt, aes(x = cluster, y = feature, fill = value)) + 
    geom_tile() + 
    scale_fill_gradientn(colors = colorRampPalette(col.scale)(50)) + 
    theme_bw() + theme(panel.grid.major = element_blank(),
                       panel.grid.minor = element_blank()) +
    theme(axis.text.x = element_text(angle = 90, vjust = 0.5, hjust=1))
```


## Smaller heatmap of marker genes
The heatmap of Figure 5F:
```{r heatmapF}
markers <- c("Lmo4",  "Cebpa",  "Cebpb",  "Cebpe",  "Spi1",
             "Csf1r", "Csf2ra", "Csf2rb", "Csf3r",  "Fcer1g",
             "Cd52",  "F7",     "F13a1",  "Ly86",   "Mpo",
             "Elane", "Fcnb",   "Ltf",    "S100a8", "S100a9", 
             "Cd74")
# focus on C1 to C8, and control and DKO only
filter <- is.element(pbmc$manual_clusters, paste0("C", 1:8)) &
          is.element(pbmc$sample, c("control", "DKO")) 
pbmc.subset <- pbmc[, filter]

# define super-clusters
pbmc.subset$super <- rep(NA, length(pbmc.subset$manual_clusters))
pbmc.subset$super[is.element(pbmc.subset$manual_clusters, paste0("C", 1:5))] <- "MPP"
pbmc.subset$super[is.element(pbmc.subset$manual_clusters, paste0("C", 6:8))] <- "ery"

# combine samples and super-clusters
pbmc.subset$super_sample <- paste(pbmc.subset$super,pbmc.subset$sample, sep = "_")


# get average expression of markers per sample in these clusters
averages <- AverageExpression(pbmc.subset, assays = "RNA", group.by = "super_sample", features = markers)
dim(averages$RNA)
averages <- averages$RNA
averages[1:5,]

# scale
averages.norm <- t(scale(t(averages)))
# melt to long form
dat.melt <- melt(averages.norm, varnames = c("feature", "cluster"))
# to reverse the order in the heatmap
dat.melt$feature <- factor(dat.melt$feature, 
                             levels = rev(rownames(averages.norm)))
ggplot(dat.melt, aes(x = cluster, y = feature, fill = value)) + 
    geom_tile() + 
    scale_fill_gradientn(colors = colorRampPalette(col.scale)(50)) + 
    theme_bw() + theme(panel.grid.major = element_blank(),
                       panel.grid.minor = element_blank()) +
    theme(axis.text.x = element_text(angle = 90, vjust = 0.5, hjust=1))

# clean up
rm(pbmc.subset)
gc()
```

## Marker genes for HSC, MPP2, MPP3, MPP4
This is to reproduce the figures in supplementary material showing the signal of marker genes for HSC, MPP2, MPP2, and MPP4 cell types. See the hsc_markers.csv, mpp2_markers.csv, mpp3_markers.csv, and mpp4_markers.csv files in this repository.

```{r otherMarkers}
# read in the markers
hsc.markers <- read.csv("hsc_markers.csv")[,2]
mpp2.markers <- read.csv("mpp2_markers.csv")[,2]
mpp3.markers <- read.csv("mpp3_markers.csv")[,2]
mpp4.markers <- read.csv("mpp4_markers.csv")[,2]

# calculate module scores
DefaultAssay(pbmc) <- 'RNA'
pbmc <- AddModuleScore(pbmc, features = list(hsc.markers), name = "hsc_")
pbmc <- AddModuleScore(pbmc, features = list(mpp2.markers), name = "mpp2_")
pbmc <- AddModuleScore(pbmc, features = list(mpp3.markers), name = "mpp3_")
pbmc <- AddModuleScore(pbmc, features = list(mpp4.markers), name = "mpp4_")
# keep in mind that AddModuleScore will append a "1" to the names...

# UMAP plots
FeaturePlot(pbmc, features = "hsc_1") + scale_colour_gradientn(colours = col.scale)
FeaturePlot(pbmc, features = "mpp2_1") + scale_colour_gradientn(colours = col.scale)
FeaturePlot(pbmc, features = "mpp3_1") + scale_colour_gradientn(colours = col.scale)
FeaturePlot(pbmc, features = "mpp4_1") + scale_colour_gradientn(colours = col.scale)

# violin plots
VlnPlot(pbmc, features = "hsc_1", group.by="manual_clusters", pt.size=0, cols=cols) + 
  theme( panel.grid.minor.y = element_line( size=.1, color="black" ),
         panel.grid.major.y = element_line( size=.1, color="black" ) ) # horizontal grid lines

VlnPlot(pbmc, features = "mpp2_1", group.by="manual_clusters", pt.size=0, cols=cols) + 
  theme( panel.grid.minor.y = element_line( size=.1, color="black" ),
         panel.grid.major.y = element_line( size=.1, color="black" ) ) # horizontal grid lines

VlnPlot(pbmc, features = "mpp3_1", group.by="manual_clusters", pt.size=0, cols=cols) + 
  theme( panel.grid.minor.y = element_line( size=.1, color="black" ),
         panel.grid.major.y = element_line( size=.1, color="black" ) ) # horizontal grid lines

VlnPlot(pbmc, features = "mpp4_1", group.by="manual_clusters", pt.size=0, cols=cols) + 
  theme( panel.grid.minor.y = element_line( size=.1, color="black" ),
         panel.grid.major.y = element_line( size=.1, color="black" ) ) # horizontal grid lines
```


## Prediction of DARs
Prediction of differentially accessible regions (DARs) between control and DKO samples, and figures 5G and 5H.

```{r DARs}
# define super-clusters
pbmc$super <- rep(NA, length(pbmc$manual_clusters))
pbmc$super[is.element(pbmc$manual_clusters, paste0("C", 1:5))] <- "MPP"
pbmc$super[is.element(pbmc$manual_clusters, paste0("C", 6:8))] <- "ery"

DefaultAssay(pbmc) <- 'peaks'

# in clusters C1 to C5 between control and DKO
da_peaks_MPP_high_in_control <- FindMarkers(
    object = subset(pbmc, super == "MPP"),
    ident.1 = "control",
    ident.2 = "DKO",
    group.by = "sample",
    min.pct = 0.1,
    logfc.threshold = 0.1,
    test.use = 'LR',
    latent.vars = 'peak_region_fragments',
    only.pos = TRUE
  )
da_peaks_MPP_high_in_DKO <- FindMarkers(
    object = subset(pbmc, super == "MPP"),
    ident.1 = "DKO",
    ident.2 = "control",
    group.by = "sample",
    min.pct = 0.1,
    logfc.threshold = 0.1,
    test.use = 'LR',
    latent.vars = 'peak_region_fragments',
    only.pos = TRUE
  )

# in clusters C6 to C8 between control and DKO
da_peaks_ery_high_in_control <- FindMarkers(
    object = subset(pbmc, super == "ery"),
    ident.1 = "control",
    ident.2 = "DKO",
    group.by = "sample",
    min.pct = 0.1,
    logfc.threshold = 0.1,
    test.use = 'LR',
    latent.vars = 'peak_region_fragments',
    only.pos = TRUE
  )
da_peaks_ery_high_in_DKO <- FindMarkers(
    object = subset(pbmc, super == "ery"),
    ident.1 = "DKO",
    ident.2 = "control",
    group.by = "sample",
    min.pct = 0.1,
    logfc.threshold = 0.1,
    test.use = 'LR',
    latent.vars = 'peak_region_fragments',
    only.pos = TRUE
  )

# choosing a set of background peaks
open.peaks <- AccessiblePeaks(pbmc)

# for each of the 4 sets of da_peaks, get enriched motifs
# note that there is a certain amount of randomness (the control regions are picked randomly)
dar_motif_prediction = function(da_peaks){
  top.da.peak <- rownames(da_peaks[da_peaks$p_val_adj < 0.0001,])
  message(length(top.da.peak))
  
  # match the overall GC content in the peak set using the open peaks set
  meta.feature <- GetAssayData(pbmc, assay = "peaks", slot = "meta.features")
  peaks.matched <- MatchRegionStats(
    meta.feature = meta.feature[open.peaks, ],
    query.feature = meta.feature[top.da.peak, ],
    n = 50000
  )
  
  # test enrichment
  enriched.motifs <- FindMotifs(
    object = pbmc,
    features = top.da.peak,
    background = peaks.matched
  )
}

motifs_MPP_high_in_control <- dar_motif_prediction(da_peaks_MPP_high_in_control)
# based on 238 DARs
motifs_MPP_high_in_DKO     <- dar_motif_prediction(da_peaks_MPP_high_in_DKO)
# based on 227 DARs
motifs_ery_high_in_control <- dar_motif_prediction(da_peaks_ery_high_in_control)
# based on 90 DARs
motifs_ery_high_in_DKO     <- dar_motif_prediction(da_peaks_ery_high_in_DKO)
# based on 246 DARs

# function for making volcano plots from pairs of the above results
make_volcano_2_sets = function(dat1, dat2, n1, n2, n3 = 50000){
  
  # put into the same order
  dat2 <- dat2[rownames(dat1),]
  
  message("n1: ", n1, " - n2: ", n2)
  # direct comparison set 1 vs set 2
  res <- matrix(NA, nrow=nrow(dat1), ncol = 7)
  rownames(res) <- rownames(dat1)
  colnames(res) <- c("with1", "without1", "with2", "without2", "freq1", "freq2", "pvalue")
  for(i in 1:nrow(dat1)){
    # get entries for Fisher's exact test
    entry1 <- dat1$observed[i]
    entry2 <- n1 - dat1$observed[i]
    entry3 <- dat2$observed[i]
    entry4 <- n2 - dat2$observed[i]
    m <- matrix(c(entry1, entry2, entry3, entry4), nrow=2)
    res.fisher <- fisher.test(m)
    pval <- res.fisher$p.value
    res[i,] <- c(entry1, entry2, entry3, entry4, (entry1+1)/(n1+1), (entry3+1)/(n2+1), pval)
  }
  
  df <- data.frame(
    log2fold = log2(res[,"freq1"]/res[,"freq2"]),
    log10pval = log10(res[,"pvalue"]),
    name = dat1$motif.name
  )
  rownames(df) <- rownames(res)

  p <- ggplot(df, aes(x=log2fold, y=log10pval)) +
    geom_point() + scale_y_continuous(trans="reverse") + 
    theme_bw()
  
  list(result = res,
       volcano = p)
}

# for Figure 5G
res.MPP <- make_volcano_2_sets(motifs_MPP_high_in_DKO, motifs_MPP_high_in_control, n1=227, n2=238)
head(res.MPP$result)
res.MPP$volcano

# for Figure 5H
res.ery <- make_volcano_2_sets(motifs_ery_high_in_DKO, motifs_ery_high_in_control, n1=246, n2=90)
head(res.ery$result)
res.ery$volcano

```