SF3B1 iCLIP analysis

Binding site clustering

Author

Affiliation

Dr. Mirko Brueggemann

Buchman Institute for Molecular Life Sciences

Published

September 18, 2023

1 Analysis Description

In this report we describe how binding sites are classified in groups based on their distance patterns. The main idea is that muliple peaks in close proximity are combined into a single region first. These regions are then fitted into a turned into smoothed coverage profiles, which can be represented as Uniform Manifold Approximation and Projection (UMAP) and further classified using DBSCAN

2 Load libraries

Show code

# genomics
library(rtracklayer)
library(GenomicRanges)
library(GenomicFeatures)
library(AnnotationDbi)
library(BindingSiteFinder)

# Data format
library(factoextra)
library(dplyr)
library(tidyr)
library(tibble)
library(forcats)

# visuals - plotting
library(ggplot2)
library(ggridges)
library(ggrastr)
library(ggpointdensity)
library(ggsci)
library(ggtext)
library(patchwork)
library(circlize)
library(viridis)
library(ggrepel)
library(ComplexHeatmap)

# visuals - format
library(kableExtra)
library(knitr)
library(gridExtra)
library(grid)

# calculation
library(matrixStats)
library(umap)
library(fpc)
library(dbscan)
library(multimode)

Show code

source("../styles.R")
source("../helper.R")

3 Prepartion of regions

At first binding sites closer to each other than 55nt are merged. Merged regions are then symmetrically extended to form 81nt wide bins.

Show code

load("/Users/mirko/Projects/sf3b1/02_markdowns/03_clean/01_bindingSites/data/bsTranscript.rda")
# Load clip data
clipFilesWt = "/Users/mirko/Projects/sf3b1/01_data_subsamp/wt/cov/replicate"
clipFilesMut = "/Users/mirko/Projects/sf3b1/01_data_subsamp/mut/cov/replicate"
clipFiles = c(clipFilesWt, clipFilesMut)
clipFiles = list.files(clipFiles, pattern = ".bw$", full.names = TRUE)
clipFilesP = clipFiles[grep(clipFiles, pattern = "Plus")]
clipFilesM = clipFiles[grep(clipFiles, pattern = "Minus")]
# Organize clip data in dataframe
colData = data.frame(
  id = c(1:5),
  condition = factor(c("WT", "WT", "WT", "WT", "WT"), levels = c("WT")),
                     clPlus = clipFilesP,
                     clMinus = clipFilesM)
# Make BindingSiteFinder object
bds = BSFDataSetFromBigWig(ranges = bsTranscript, meta = colData)

3.1 Distance pattern for binding sites in introns

Show code

bsIntron = subset(bsTranscript, bsTranscript$region == "intron")

dist = distanceToNearest(bsIntron) %>% as.data.frame()
bsIntron$dist = dist$distance
ggplot(dist, aes(x = log10(distance+1))) + 
  geom_histogram(bins = 100, color = "black") + 
  theme_nice() +
  labs(
    title = "Distance to nearest binding site",
    x = "Distance +1 (nt) [log10]",
    y = "Count")

Distance from each binding site to the next closest neighbor.

Show code

ggplot(dist, aes(x = distance)) + 
  geom_histogram(binwidth = 1, color = "black") + 
  xlim(-1,50) +
  theme_nice() +
  labs(
    title = "Distance to nearest binding site",
    x = "Distance (nt) [0-50]",
    y = "Count") +
  geom_vline(xintercept = 7, linetype = "dashed")

Distance from each binding site to the next closest neighbor in a range of 50 nt.

3.2 Transform the iCLIP signal

To avoid that the height of the coverage influences our downstream clustering, the signal in each bin is scaled between 0 and 1. Next a spline transformation is applied to produced a smoothed version of the coverage, which boosts classification performance. After testing different lambda values (0.2, 0.3, 0.4) we decided from the heatmaps below to continue with lambda = 0.2 and a dimension dim = 150.

Show code

# combine all BS within 41 nt range
mergedRange = reduce(bsIntron, min.gapwidth = 55, with.revmap = TRUE)
mergedRange$width = width(mergedRange)
# add bs count info 
nBS = sapply(mergedRange$revmap, length)
mcols(mergedRange)$nBS = nBS

# resize ranges to their center position and extend all ranges to 81 nt
rngSel = resize(granges(mergedRange), fix = "center", width = 81)
names(rngSel) = 1:length(rngSel)

export(mergedRange, "./data/mergedRange.bed", format = "BED")
export(rngSel, "./data/rngSel.bed", format = "BED")

saveRDS(mergedRange, file = "./data/mergedRange.rds")
saveRDS(rngSel, file = "./data/rngSel.rds")

Show code

bdsSel = setRanges(bds, rngSel)
cov = coverageOverRanges(bdsSel, returnOptions = "merge_all_replicates")
masterMM = cov

# normalize matrix
normMM = minMaxNorm(masterMM)

# # apply different smoothing levels
smoothMM_par1 <- t(apply(normMM, 1, smoothing, lambda=0.2, dim=151))
smoothMM_par2 <- t(apply(normMM, 1, smoothing, lambda=0.3, dim=151))
smoothMM_par3 <- t(apply(normMM, 1, smoothing, lambda=0.4, dim=151))
saveRDS(smoothMM_par1, file = "./data/smoothMM_par1.rds")
saveRDS(smoothMM_par2, file = "./data/smoothMM_par2.rds")
saveRDS(smoothMM_par3, file = "./data/smoothMM_par3.rds")

3.3 Smooting levels

Show code

set.seed(1234)
custom.col = viridis(10, option = "G", direction = -1)

idx = sample(1:(nrow(smoothMM_par1)), 500)

m1 = smoothMM_par1[idx,]
m2 = smoothMM_par2[idx,]
m3 = smoothMM_par3[idx,]

h1 = Heatmap(m1, column_title = "Smooth 0.2", name = "Xlinks",
             cluster_columns = F, cluster_rows = T,
             show_row_names = FALSE,
             show_column_names = T, border = T, col = custom.col,
             heatmap_legend_param = list(
               at = c(0,0.5, 1), labels = c("0", "0.5", "1"),
               legend_width = unit(6, "cm"),
               title_position = "topleft", direction = "horizontal"
             ), use_raster = TRUE
)
draw(h1, heatmap_legend_side = "bottom")

Heatmaps of smoothed binding site signal with level 0.2

Show code

h2 = Heatmap(m2, column_title = "Smooth 0.3", name = "Xlinks",
             cluster_columns = F, cluster_rows = T,
             show_row_names = FALSE,
             show_column_names = T, border = T, col = custom.col,
             heatmap_legend_param = list(
               at = c(0,0.5, 1), labels = c("0", "0.5", "1"),
               legend_width = unit(6, "cm"),
               title_position = "topleft", direction = "horizontal"
             ), use_raster = TRUE
)
draw(h2, heatmap_legend_side = "bottom")

Heatmaps of smoothed binding site signal with level 0.3

Show code

h3 = Heatmap(m3, column_title = "Smooth 0.4", name = "Xlinks",
             cluster_columns = F, cluster_rows = T,
             show_row_names = FALSE,
             show_column_names = T, border = T, col = custom.col,
             heatmap_legend_param = list(
               at = c(0,0.5, 1), labels = c("0", "0.5", "1"),
               legend_width = unit(6, "cm"),
               title_position = "topleft", direction = "horizontal"
             ), use_raster = TRUE
)
draw(h3, heatmap_legend_side = "bottom")

4 UMAP classification of intronic binding regions

UMAP classification is calculated with settings as proposed in the STOATY dive approach for coverage shape based binding site clustering. Setting details are: n_epochs = 5000, n_components = 2, min_dist = 0.01 and n_neighbors = 5.

Show code

umapDf_par1 = readRDS("./data/umapDf_par1.rds")

5 Density based clustering (DBSCAN) of UMAP result

UMAP results are clustered with DBSCAN (density based clustering). This method defines cluster on the observed density, based on the definition of cluster centers. In particular the minimal number of points that are needed to form a cluster and the distance that points need to have to be assigned to a cluster. The following settings were used: eps = 0.3 and MinPts = 150.

5.1 DBSCAN results

Show code

set.seed(1234)
dbscan::kNNdistplot(umapDf_par1$layout, k = 150)
abline(h = 0.3, lty = 2)

K nearest neighbors plot with k 150. K equals the MinPts option in the dbscan. Line at 0.3 represents eps param.

Show code

grouping = dbscan::dbscan(umapDf_par1$layout, eps = 0.3, minPts = 150)
grouping$cluster = ifelse(grouping$cluster == 2, 3, ifelse(grouping$cluster == 3, 2, grouping$cluster))

df1 = data.frame(x = umapDf_par1$layout[,1],
                y = umapDf_par1$layout[,2],
                group = factor(grouping$cluster))

df1 = subset(df1, group != 0)

ggplot(df1, aes(x, y)) +
  ggrastr::rasterise(geom_pointdensity(size = 0.2), dpi = 300) +
  theme_pub() +
  scale_color_viridis(option = "A") +
  theme(legend.position = "top",  aspect.ratio = 1) +
  ggforce::geom_mark_ellipse(aes(label = group, group = group), alpha = 0, label.fontsize = 12, expand = unit(2, "mm")) +
  guides(color = guide_colorbar(title.position = 'top', title.hjust = 0.5,
                                barwidth = unit(20, 'lines'), barheight = unit(.5, 'lines'))) +
  scale_y_continuous(limits = c(-12, 20)) +
  scale_x_continuous(limits = c(-12, 8)) +
  labs(
    x = "UMAP 1",
    y = "UMAP 2",
  )

DBSCAN clustering of UMAP transformation for regions with smooting 0.2 and dbscan clustering; Outlier cluster 0 removed.

5.2 Cluster examples as heatmap

Show code

set.seed(1234)

s = split.data.frame(smoothMM_par1, grouping$cluster)
# s$rest = rbind(s$`1`, s$`2`, s$`3`)
s = lapply(s, head, n = 300)

custom.col = viridis(10, option = "G", direction = -1)
clustRow = FALSE

# annotate with top col sums profile m1
df1 = data.frame(sums = colMeans(s$`1`))
haMeans1 = HeatmapAnnotation(cov = anno_barplot(df1, gp = gpar(fill = "#595959", col = "#595959")),
                            height = unit(2, "cm"), show_annotation_name = F)
# annotate with top col sums profile m2 
df2 = data.frame(sums = colMeans(s$`2`))
haMeans2 = HeatmapAnnotation(cov = anno_barplot(df2, gp = gpar(fill = "#595959", col = "#595959")),
                            height = unit(2, "cm"), show_annotation_name = F)
# annotate with top col sums profile m3
df3 = data.frame(sums = colMeans(s$`3`))
haMeans3 = HeatmapAnnotation(cov = anno_barplot(df3, gp = gpar(fill = "#595959", col = "#595959")),
                            height = unit(2, "cm"), show_annotation_name = F)

h1 = Heatmap(s$`1`, column_title = "Cluster 1", name = "Xlinks",
             cluster_rows = clustRow, cluster_columns = F, show_row_names = FALSE,
             show_column_names = T, border = T, col = custom.col,
             top_annotation = haMeans1,
             heatmap_legend_param = list(
               at = c(0,0.5, 1), labels = c("0", "0.5", "1"),
               legend_width = unit(6, "cm"),
               title_position = "topleft", direction = "horizontal"
             ), use_raster = TRUE
)
h2 = Heatmap(s$`2`, column_title = "Cluster 2", name = "Xlinks",
        cluster_rows = clustRow, cluster_columns = F, show_row_names = FALSE,
        top_annotation = haMeans2,
        show_column_names = T, border = T, col = custom.col, use_raster = TRUE
)
h3 = Heatmap(s$`3`, column_title = "Cluster 3", name = "Xlinks",
        cluster_rows = clustRow, cluster_columns = F, show_row_names = FALSE,
        top_annotation = haMeans3,
        show_column_names = T, border = T, col = custom.col, use_raster = TRUE
)

l = h1 + h2 + h3
draw(l, heatmap_legend_side = "bottom")

moothed crosslink heatmaps split by kmeans clustering. Subsetted to top (head) 300.

Show code

mcols(rngSel)$MajorCluster = grouping$cluster
rngExport = rngSel
names(rngExport) = ifelse(rngExport$MajorCluster == 1, paste0("SinglePeak_", names(rngExport)),
                          ifelse(rngExport$MajorCluster == 2, paste0("DoubleNarrow_", names(rngExport)),
                                 ifelse(rngExport$MajorCluster == 3, paste0("DoubleWide_", names(rngExport)), paste0("Rest_",names(rngExport)))))
mcols(rngExport)$itemRgb = ifelse(rngExport$MajorCluster == 1, viridis(option = "G", n = 5)[2], 
                                  ifelse(rngExport$MajorCluster == 2, viridis(option = "G", n = 5)[3],
                                         ifelse(rngExport$MajorCluster == 3, viridis(option = "G", n = 5)[4], "Grey")))
rtracklayer::export(rngExport, "./data/rngClassified.bed", format = "BED",
                    trackLine = new("BasicTrackLine", name="Ranges major cluster", itemRgb = TRUE)
                    )

6 Features of the clustered groups

6.1 Explorative plots for result clusters.

Show code

s = split.data.frame(smoothMM_par1, grouping$cluster)
df = data.frame(cluster = factor(names(s)), size = sapply(s, nrow))

p0 = ggplot(df, aes(x = cluster, y = size)) +
  geom_col() +
  theme_pub() +
  labs(y = "N") +
  scale_y_log10() +
  geom_text(aes(label = size), vjust = 1.3, color = "white") +
  labs(
    title = "Ranges per group",
    x = "Cluster",
    y = "Count"
  )
names(mergedRange) = 1:length(mergedRange)
rngC1 = mergedRange[names(mergedRange) %in% rownames(s$`1`)]
rngC2 = mergedRange[names(mergedRange) %in% rownames(s$`2`)]
rngC3 = mergedRange[names(mergedRange) %in% rownames(s$`3`)]
bsC1 = subsetByOverlaps(bsIntron, rngC1)
bsC2 = subsetByOverlaps(bsIntron, rngC2)
bsC3 = subsetByOverlaps(bsIntron, rngC3)

df = data.frame(cluster = rep(c("C1", "C2", "C3"),2),
                type = c(rep("BS",3),rep("Region",3)),
                value = c(length(bsC1), length(bsC2), length(bsC3),
                          length(rngC1), length(rngC2), length(rngC3))
                )
p1 = ggplot(df, aes(x = cluster, y = value, fill = type)) +
  geom_col(position = "fill") +
  theme_pub() +
  theme(legend.position = "top") +
  scale_fill_npg() +
  labs(
    x = "Cluster",
    y = "Counts",
    fill = "Type"
  )

p2 = ggplot(df, aes(x = cluster, y = value, fill = type)) +
  geom_col(position = "dodge") +
  theme_pub() +
  theme(legend.position = "top") +
  scale_fill_npg() +
  labs(
    x = "Cluster",
    y = "Counts",
    fill = "Type"
  )

# names(mergedRange) = 1:length(mergedRange)
# s = split.data.frame(smoothMM_par1, grouping$cluster)
df1 = data.frame(N = mergedRange$nBS[names(mergedRange) %in% rownames(s$`1`)], cluster = "C1")
df2 = data.frame(N = mergedRange$nBS[names(mergedRange) %in% rownames(s$`2`)], cluster = "C2")
df3 = data.frame(N = mergedRange$nBS[names(mergedRange) %in% rownames(s$`3`)], cluster = "C3")
df = rbind(df1,df2,df3)
df$N[df$N > 7] = 7
df = table(df$N, df$cluster) %>% as.data.frame()

p3 = ggplot(df, aes(x = Var1, y = Freq+1, fill = Var2)) +
  geom_col(position = "dodge") +
  scale_y_log10() +
  theme_pub() +
  scale_fill_npg() +
  theme(legend.position = "top") +
  labs(
    title = "Binding sites per group and cluster",
    x = "Number of Binding sites",
    y = "Count",
    fill = "Cluster"
  )

p4 = ggplot(df, aes(x = Var1, y = Freq+1, fill = Var2)) +
  geom_col(position = "fill") +
  # scale_y_log10() +
  theme_pub() +
  scale_fill_npg() +
  theme(legend.position = "top") +
  labs(
    title = "Binding sites per group and cluster",
    x = "Number of Binding sites",
    y = "Count",
    fill = "Cluster"
  )

p5 = ggplot(df, aes(x = Var1, y = Freq+1, fill = Var2)) +
  geom_col(position = "dodge") +
  scale_y_log10() +
  theme_pub() +
  scale_fill_npg() +
  theme(legend.position = "top") +
  labs(
    title = "Binding sites per group and cluster",
    x = "Number of Binding sites",
    y = "Count",
    fill = "Cluster"
  ) +
  facet_grid(~Var2)

(p2 + p1) / (p0 + p3) / (p4 + p5)

Number of binding sites in merged regions and clusters.} Binding site overlaps counted in merged ranges (sizes is variable here).

Show code

s = split.data.frame(smoothMM_par1, grouping$cluster)
df = data.frame(cluster = factor(names(s)), size = sapply(s, nrow))

p0 = ggplot(df, aes(x = cluster, y = size)) +
  geom_col() +
  theme_pub() +
  labs(y = "N") +
  scale_y_log10() +
  geom_text(aes(label = size), vjust = 1.3, color = "white") +
  labs(
    title = "Ranges per group",
    x = "Cluster",
    y = "Count"
  )
rngC1 = rngSel[names(rngSel) %in% rownames(s$`1`)]
rngC2 = rngSel[names(rngSel) %in% rownames(s$`2`)]
rngC3 = rngSel[names(rngSel) %in% rownames(s$`3`)]
bsC1 = subsetByOverlaps(bsIntron, rngC1)
bsC2 = subsetByOverlaps(bsIntron, rngC2)
bsC3 = subsetByOverlaps(bsIntron, rngC3)

df = data.frame(cluster = rep(c("C1", "C2", "C3"),2),
                type = c(rep("BS",3),rep("Region",3)),
                value = c(length(bsC1), length(bsC2), length(bsC3),
                          length(rngC1), length(rngC2), length(rngC3))
                )
p1 = ggplot(df, aes(x = cluster, y = value, fill = type)) +
  geom_col(position = "fill") +
  theme_pub() +
  theme(legend.position = "top") +
  scale_fill_npg() +
  labs(
    x = "Cluster",
    y = "Counts",
    fill = "Type"
  )

p2 = ggplot(df, aes(x = cluster, y = value, fill = type)) +
  geom_col(position = "dodge") +
  theme_pub() +
  theme(legend.position = "top") +
  scale_fill_npg() +
  labs(
    x = "Cluster",
    y = "Counts",
    fill = "Type"
  )

df1 = data.frame(N = countOverlaps(rngC1, bsIntron), cluster = "C1")
df2 = data.frame(N = countOverlaps(rngC2, bsIntron), cluster = "C2")
df3 = data.frame(N = countOverlaps(rngC3, bsIntron), cluster = "C3")
df = rbind(df1,df2,df3)
df$N[df$N > 5] = 5
df = table(df$N, df$cluster) %>% as.data.frame()

p3 = ggplot(df, aes(x = Var1, y = Freq+1, fill = Var2)) +
  geom_col(position = "dodge") +
  scale_y_log10() +
  theme_pub() +
  scale_fill_npg() +
  theme(legend.position = "top") +
  labs(
    title = "Binding sites per group and cluster",
    x = "Number of Binding sites",
    y = "Count",
    fill = "Cluster"
  ) +
  geom_text(aes(label = Freq), vjust = 0.5, hjust = 1.2, position = position_dodge(width = .9), angle = 90) 

p4 = ggplot(df, aes(x = Var1, y = Freq+1, fill = Var2)) +
  geom_col(position = "fill") +
  # scale_y_log10() +
  theme_pub() +
  scale_fill_npg() +
  theme(legend.position = "top") +
  labs(
    title = "Binding sites per group and cluster",
    x = "Number of Binding sites",
    y = "Count",
    fill = "Cluster"
  ) +
  scale_x_discrete(breaks=c("1", "2", "3", "4", "5"),
        labels=c("1", "2", "3", "4", "5+"))

p5 = ggplot(df, aes(x = Var1, y = Freq+1, fill = Var2)) +
  geom_col(position = "dodge") +
  scale_y_log10() +
  theme_pub() +
  scale_fill_npg() +
  theme(legend.position = "top") +
  labs(
    title = "Binding sites per group and cluster",
    x = "Number of Binding sites",
    y = "Count",
    fill = "Cluster"
  ) +
  facet_grid(~Var2)

(p2 + p1) / (p0 + p3) / (p4 + p5)

Number of binding sites in merged regions and clusters. Binding site overlaps counted in 81nt clustered ranges.

6.2 Binding sites per cluster selection

Show code

df2 = df
df2$Var1 = ifelse(df2$Var1 == 5, "5+", df$Var1)

p1 = ggplot(df2, aes(x = Var2, y = Freq+1, fill = Var1)) +
  geom_col(position = "fill") +
  theme_pub() +
  scale_fill_npg() +
  theme(legend.position = "top") +
  labs(
    title = "Number of binding sites per cluster",
    x = "Cluster",
    y = "Fraction",
    fill = "Binding sites"
  ) +
  theme(aspect.ratio = 1)
p1

Show code

p2 = ggplot(df2, aes(x = Var2, y = Freq+1, fill = Var1)) +
  geom_col(position = "fill") +
  theme_pub() +
  scale_fill_grey() +
  theme(legend.position = "top") +
  labs(
    title = "Number of binding sites per cluster",
    x = "Cluster",
    y = "Fraction",
    fill = "Binding sites"
  ) +
  theme(aspect.ratio = 1)
p2

Show code

p3 = ggplot(df2, aes(x = Var2, y = Freq+1, fill = Var1)) +
  geom_col(position = "fill") +
  theme_pub() +
  scale_fill_brewer(palette = 1, direction = 1) +
  theme(legend.position = "top") +
  labs(
    title = "Number of binding sites per cluster",
    x = "Cluster",
    y = "Fraction",
    fill = "Binding sites"
  ) +
  theme(aspect.ratio = 1)
p3

Show code

df1 = data.frame(N = countOverlaps(rngC1, bsIntron), cluster = "C1")
df2 = data.frame(N = countOverlaps(rngC2, bsIntron), cluster = "C2")
df3 = data.frame(N = countOverlaps(rngC3, bsIntron), cluster = "C3")
df = rbind(df1,df2,df3)

d = df %>% group_by(cluster) %>% summarize(mean = mean(N), median = median(N)) %>% as.data.frame()

kable(d, caption = "Number of BS per cluster") %>% 
  kable_styling("striped") %>%
  scroll_box(width = "100%")

Number of BS per cluster
cluster	mean	median
C1	1.057454	1
C2	2.081810	2
C3	2.615946	2

6.3 Distance to splice site per cluster

Show code

load("/Users/mirko/Projects/Annotations/human/gencode_36/filtered/gencode_v36_filtered.rda")
anno.db = loadDb("/Users/mirko/Projects/Annotations/human/gencode_36/filtered/gencode_v36_filtered.sqlite")
gns = genes(anno.db)
idx = match(gns$gene_id, anno$gene_id)
elementMetadata(gns) = cbind(elementMetadata(gns), elementMetadata(anno)[idx,])

Show code

x0 = subset(anno, transcript_type == "protein_coding")
x1 = subset(x0, type == "exon" | type == "CDS")
x2 = subset(x1, tag == "exp_conf" | tag == "CCDS" | tag == "basic")

exn = exons(anno.db)
exn = subsetByOverlaps(exn, x2, type = "within")
export(granges(exn), "./data/exn.bed", format = "BED")

spliceSites3 = unique(resize(exn, fix = "start", width = 1)) 
spliceSites5 = unique(resize(exn, fix = "end", width = 1)) 

calcDist <- function(rng) {
  # identify the next downstream 3'SS for each given range
  idx3 = precede(rng, spliceSites3)
  # identify the next downstream 3'SS for each given range
  idx5 = follow(rng, spliceSites5)
  
  if (any(is.na(idx3), is.na(idx5))) {
    removeIdx3 = which(is.na(idx3))
    removeIdx5 = which(is.na(idx5))
    removeIdx = c(removeIdx3, removeIdx5)
    idx3 = idx3[-removeIdx]
    idx5 = idx5[-removeIdx]
    rng = rng[-removeIdx]
  }
  
  # calculate distance between given range and associated splice site
  d1 = distance(rng, spliceSites3[idx3])
  d2 = distance(rng, spliceSites5[idx5])
  
  df = data.frame(d3ss = d1, d5ss = d2)
  return(df)
}

Show code

s = split.data.frame(smoothMM_par1, grouping$cluster)
originalRange = mergedRange
originalRange = granges(originalRange)
names(originalRange) = names(rngSel)
rngC1 = originalRange[names(originalRange) %in% rownames(s$`1`)]
rngC2 = originalRange[names(originalRange) %in% rownames(s$`2`)]
rngC3 = originalRange[names(originalRange) %in% rownames(s$`3`)]

# calculate distance
df1 = calcDist(rngC1) %>% 
    as.data.frame() %>%
    mutate(clust = "C1: single") %>%
    pivot_longer(-clust)

df2 = calcDist(rngC2) %>% 
    as.data.frame() %>%
    mutate(clust = "C2: double-narrow") %>%
    pivot_longer(-clust)

df3 = calcDist(rngC3) %>% 
    as.data.frame() %>%
    mutate(clust = "C3: double-wide") %>%
    pivot_longer(-clust)

df = rbind(df1,df2,df3)
df$value[df$value == 0] = 1

p1 = ggplot(df, aes(x = clust, y = log10(value), fill = name)) +
    geom_boxplot(outlier.shape = NA) +
    theme_pub() +
    scale_fill_npg() +
    labs(
        x = "Cluster",
        y = "Distance to splice site (log10)",
        fill = "Splice site"
    )
p1

Distances to splice sites. Distance is calculated from the outer edge of the outer binding sites for each region to the respective 3’ss and 5’ss splice site”

Show code

df1 = calcDist(rngC1) %>% 
    as.data.frame() %>%
    mutate(clust = "C1: single", r = d3ss/ d5ss) 

df2 = calcDist(rngC2) %>% 
    as.data.frame() %>%
    mutate(clust = "C2: double-narrow", r = d3ss/ d5ss) 

df3 = calcDist(rngC3) %>% 
    as.data.frame() %>%
    mutate(clust = "C3: double-wide", r = d3ss/ d5ss) 

df = rbind(df1,df2,df3)

p2 = ggplot(df, aes(x = clust, y = log2(r), fill = clust)) +
    geom_boxplot(outlier.shape = NA) +
    theme_pub() +
    scale_fill_manual(values = viridis(n = 10, option = "mako")[c(3,5,7)]) +
    theme(legend.position = "none") +
    labs(
        x = "Cluster",
        y = "Splice site distance ratio (log2(3'ss/5'ss))"
    )
p2

7 UMAP subclassification of major cluster 3 - double-wide pattern

To further refine the positioning of double-peaks within the cluster 3 groups, we performed a second round of UMAP + DBSCAN. After cleaning resulting clusters each cluster gets assigned to a group based on the distance of the modes between both double-peaks.

Show code

selMM_g3 = masterMM[grouping$cluster == 3,]

normMM_g3 = apply(selMM_g3, 1, function(x){
  n = ((x - min(x)) / (max(x) - min(x)))
  return(n)
})
normMM_g3 = t(normMM_g3)

smoothMM_g3_par1 <- t(apply(normMM_g3, 1, smoothing, lambda=0.1, dim=500))
saveRDS(smoothMM_g3_par1, "./data/smoothMM_g3_par1.rds")

Show code

umapDf_g3_par1 = readRDS("./data/umapDf_g3_par1.rds")

7.1 DBSCAN clustering results

Show code

set.seed(1234)
dbscan::kNNdistplot(umapDf_g3_par1$layout, k = 60)
abline(h = 0.22, lty = 2)

Show code

groupingG3 = dbscan::dbscan(umapDf_g3_par1$layout, eps = 0.23, MinPts = 60)

df1 = data.frame(x = umapDf_g3_par1$layout[,1],
                y = umapDf_g3_par1$layout[,2],
                group = factor(groupingG3$cluster))

df1 = df1 %>% subset(group != "0")
p3 = ggplot(df1, aes(x, y, color = group)) +
  # geom_point(size = 0.2) +
  ggrastr::geom_point_rast(size = 0.2) +
  theme_pub() +
  # scale_color_manual(values = viridis(n = 6, option = "mako")[2:8]) +
  guides(colour = guide_legend(override.aes = list(size=2), nrow = 4)) +
  theme(legend.position = "none") +
  labs(title = "Smooth 0.1", 
       color = "C")

p3

7.2 Order cluster groups by peak mode distance

Show code

set.seed(1234)

# split smoothed coverage matrix by clustering groups
s = split.data.frame(smoothMM_g3_par1, groupingG3$cluster)
tmp = lapply(1:length(s), function(x){
    z = floor(ncol(s[[1]]) / 2)
    d = data.frame(Group = factor(rep(x, ncol(s[[x]]))),
                   value = colMeans(s[[x]], na.rm = TRUE),
                   pos = -246:247)
    
})
df = dplyr::bind_rows(tmp, .id = "variable")
df$Group = as.factor(as.numeric(df$Group)-1)

# calc new position
dimOrg = 81
dimSmooth = 500
dimSf = dimSmooth / dimOrg
df$newPos = df$pos / dimSf

# split by group
df = df[df$Group != 0,]
dfSplit = split(df, df$Group)
dfSplit = lapply(dfSplit, function(x){
    data.frame(v = rep(x$newPos, ifelse(x$value < 0, x$value*(-1000), x$value*1000)))
})
dfSplit = dplyr::bind_rows(dfSplit, .id = "variable")
dfSplit = dfSplit[dfSplit$variable != 0,]

# calculate modes
modeDf = split.data.frame(dfSplit, dfSplit$variable)
modes = lapply(modeDf, function(x){
    m = locmodes(x$v, mod0 = 2)
    d = m$locations
    return(d)
})
modes = dplyr::bind_rows(modes, .id = "variable") %>% 
    t() %>% 
    as.data.frame() %>% 
    tibble::rownames_to_column("cluster") %>% 
    rename(mode1 = V1, mode2 = V3, antimode = V2)

dfLine = modes %>%
    select(cluster, mode1, mode2) %>%
    pivot_longer(-cluster) %>%
    rename(variable = cluster)

dist = length(modes$mode1:modes$mode2)

dfDist = modes %>%
    group_by(cluster) %>%
    mutate(dist = length(mode1:mode2)) %>%
    select(cluster, dist) %>%
    rename(variable = cluster)

fOrder = dfDist$variable[order(dfDist$dist)]
dfSplit$variable = factor(dfSplit$variable, levels = fOrder)
dfLine$variable = factor(dfLine$variable, levels = fOrder)
dfDist$variable = factor(dfDist$variable, levels = fOrder)

dfOrder = dfDist[order(dfDist$dist),]
dfOrder$NewCluster = 1:nrow(dfOrder)

idx = match(df1$group, dfOrder$variable)
df1$NewCluster = as.factor(dfOrder$NewCluster[idx])

ggplot(subset(df1, NewCluster != 0), aes(x, y, fill = NewCluster)) +
    rasterize(geom_point(shape = 21, stroke = 0.3), dpi = 300) +
    theme_pub() +
    scale_fill_viridis(option = "turbo", discrete = T, direction = -1) +
    theme(legend.position = "none", aspect.ratio = 1) +
    scale_y_continuous(limits = c(-10, 10)) + 
    scale_x_continuous(limits = c(-10, 15)) +
    labs(
        x = "UMAP 1",
        y = "UMAP 2",
    ) +
    ggforce::geom_mark_ellipse(aes(label = NewCluster, group = NewCluster),
                               alpha = 0, label.fontsize = 12,
                               label.buffer = unit(5, 'mm'), expand = unit(2, "mm"),
                               label.minwidth = unit(5, "mm"))

Sub-cluster 3 groups matched by double peak mode distances

Show code

df = dfSplit

dfOrder = dfDist[order(dfDist$dist),]
dfOrder$NewCluster = 1:nrow(dfOrder)

idx = match(df$variable, dfOrder$variable)
df$NewCluster = dfOrder$NewCluster[idx]
df$name = paste0("cluster: ", df$NewCluster, ", distance: ", dfOrder$dist[idx])

df$name = factor(df$name)
df$name = factor(df$name, levels = as.character(unique(df$name)[as.numeric(levels(df$variable))]))

p = ggplot(df, aes(x = v, y = name, fill = ..density..)) +
  rasterise(geom_density_ridges_gradient(scale = 10, bandwidth = 0.1, color = "black", size = 0.5),dpi = 300) +
  # theme_pub() +
  theme_pub() +
  theme(legend.position = "top") +
  scale_y_discrete(expand = expand_scale(mult = c(0, 0.15))) +
  scale_fill_gradient2(position="top" , low = "white", high = "#366A9FFF", mid = "348AA6FF", midpoint = 0.05, breaks = c(0.01, 0.05, 0.1)) +
  labs(
    title = "Sub-cluster by double-mode distance",
    x = "Relative distance (nt)",
    y = "Cluster"
  )
p

Sub-cluster 3 smoothed crosslink densities split by density based clustering.} Sorted by double-mode distances.

8 Session Information

Show code

sessionInfo()

R version 4.2.1 (2022-06-23)
Platform: x86_64-apple-darwin17.0 (64-bit)
Running under: macOS Big Sur ... 10.16

Matrix products: default
BLAS:   /Library/Frameworks/R.framework/Versions/4.2/Resources/lib/libRblas.0.dylib
LAPACK: /Library/Frameworks/R.framework/Versions/4.2/Resources/lib/libRlapack.dylib

locale:
[1] en_US.UTF-8/en_US.UTF-8/en_US.UTF-8/C/en_US.UTF-8/en_US.UTF-8

attached base packages:
[1] grid      stats4    stats     graphics  grDevices utils     datasets 
[8] methods   base     

other attached packages:
 [1] multimode_1.5           dbscan_1.1-11           fpc_2.2-10             
 [4] umap_0.2.10.0           matrixStats_1.0.0       gridExtra_2.3          
 [7] knitr_1.43              kableExtra_1.3.4        ComplexHeatmap_2.14.0  
[10] ggrepel_0.9.3           viridis_0.6.3           viridisLite_0.4.2      
[13] circlize_0.4.15         patchwork_1.1.2         ggtext_0.1.2           
[16] ggsci_3.0.0             ggpointdensity_0.1.0    ggrastr_1.0.2          
[19] ggridges_0.5.4          forcats_1.0.0           tibble_3.2.1           
[22] tidyr_1.3.0             dplyr_1.1.2             factoextra_1.0.7       
[25] ggplot2_3.4.2           BindingSiteFinder_1.7.8 GenomicFeatures_1.49.7 
[28] AnnotationDbi_1.59.1    Biobase_2.57.1          rtracklayer_1.57.0     
[31] GenomicRanges_1.49.1    GenomeInfoDb_1.33.10    IRanges_2.31.2         
[34] S4Vectors_0.35.4        BiocGenerics_0.43.4    

loaded via a namespace (and not attached):
  [1] BiocFileCache_2.5.2         systemfonts_1.0.4          
  [3] plyr_1.8.8                  BiocParallel_1.31.13       
  [5] digest_0.6.31               foreach_1.5.2              
  [7] htmltools_0.5.5             magick_2.7.4               
  [9] fansi_1.0.4                 magrittr_2.0.3             
 [11] memoise_2.0.1               cluster_2.1.4              
 [13] doParallel_1.0.17           ks_1.14.0                  
 [15] Biostrings_2.65.6           svglite_2.1.1              
 [17] askpass_1.1                 prettyunits_1.1.1          
 [19] colorspace_2.1-0            blob_1.2.4                 
 [21] rvest_1.0.3                 rappdirs_0.3.3             
 [23] ggdist_3.3.0                xfun_0.39                  
 [25] crayon_1.5.2                RCurl_1.98-1.12            
 [27] jsonlite_1.8.5              iterators_1.0.14           
 [29] glue_1.6.2                  polyclip_1.10-4            
 [31] gtable_0.3.3                zlibbioc_1.43.0            
 [33] XVector_0.37.1              webshot_0.5.4              
 [35] GetoptLong_1.0.5            DelayedArray_0.23.2        
 [37] distributional_0.3.2        kernlab_0.9-32             
 [39] shape_1.4.6                 DEoptimR_1.1-2             
 [41] prabclus_2.3-2              scales_1.2.1               
 [43] mvtnorm_1.2-2               DBI_1.1.3                  
 [45] Rcpp_1.0.10                 progress_1.2.2             
 [47] gridtext_0.1.5              clue_0.3-64                
 [49] reticulate_1.30             mclust_6.0.0               
 [51] bit_4.0.5                   htmlwidgets_1.6.2          
 [53] httr_1.4.6                  RColorBrewer_1.1-3         
 [55] modeltools_0.2-23           flexmix_2.3-19             
 [57] pkgconfig_2.0.3             XML_3.99-0.14              
 [59] farver_2.1.1                nnet_7.3-19                
 [61] dbplyr_2.3.2                utf8_1.2.3                 
 [63] labeling_0.4.2              tidyselect_1.2.0           
 [65] rlang_1.1.1                 munsell_0.5.0              
 [67] tools_4.2.1                 cachem_1.0.8               
 [69] cli_3.6.1                   generics_0.1.3             
 [71] RSQLite_2.3.1               evaluate_0.21              
 [73] stringr_1.5.0               fastmap_1.1.1              
 [75] yaml_2.3.7                  bit64_4.0.5                
 [77] robustbase_0.99-0           purrr_1.0.1                
 [79] KEGGREST_1.37.3             rootSolve_1.8.2.3          
 [81] pracma_2.4.2                xml2_1.3.4                 
 [83] biomaRt_2.53.3              compiler_4.2.1             
 [85] rstudioapi_0.14             beeswarm_0.4.0             
 [87] filelock_1.0.2              curl_5.0.1                 
 [89] png_0.1-8                   tweenr_2.0.2               
 [91] stringi_1.7.12              highr_0.10                 
 [93] RSpectra_0.16-1             lattice_0.21-8             
 [95] Matrix_1.5-4.1              vctrs_0.6.3                
 [97] pillar_1.9.0                lifecycle_1.0.3            
 [99] GlobalOptions_0.1.2         bitops_1.0-7               
[101] R6_2.5.1                    BiocIO_1.7.1               
[103] KernSmooth_2.23-21          vipor_0.4.5                
[105] codetools_0.2-19            MASS_7.3-60                
[107] SummarizedExperiment_1.27.3 openssl_2.0.6              
[109] rjson_0.2.21                withr_2.5.0                
[111] GenomicAlignments_1.33.1    Rsamtools_2.13.4           
[113] GenomeInfoDbData_1.2.9      diptest_0.76-0             
[115] parallel_4.2.1              hms_1.1.3                  
[117] class_7.3-22                rmarkdown_2.22             
[119] MatrixGenerics_1.9.1        Cairo_1.6-0                
[121] ggforce_0.4.1               ggbeeswarm_0.7.2           
[123] restfulr_0.0.15

--- title: "SF3B1 iCLIP analysis" subtitle: "Binding site clustering" date: "`r format(Sys.time(), '%B %e, %Y')`" author: - name: "Dr. Mirko Brueggemann" email: mirko.brueggemann@bmls.de affiliations: - name: Buchman Institute for Molecular Life Sciences format: html: theme: sandstone code-fold: TRUE code-overflow: scroll code-summary: "Show code" code-tools: TRUE toc: TRUE toc-depth: 3 toc-location: left number-sections: TRUE self-contained: TRUE fontsize: 11pt crossref: fig-title: '**Figure**' fig-labels: arabic title-delim: "**.**" code-block-bg: "#EEEEEE" editor: markdown: wrap: 120 --- # Analysis Description In this report we describe how binding sites are classified in groups based on their distance patterns. The main idea is that muliple peaks in close proximity are combined into a single region first. These regions are then fitted into a turned into smoothed coverage profiles, which can be represented as Uniform Manifold Approximation and Projection (UMAP) and further classified using DBSCAN # Load libraries ```{r} #| label: libraries #| message: false # genomics library(rtracklayer) library(GenomicRanges) library(GenomicFeatures) library(AnnotationDbi) library(BindingSiteFinder) # Data format library(factoextra) library(dplyr) library(tidyr) library(tibble) library(forcats) # visuals - plotting library(ggplot2) library(ggridges) library(ggrastr) library(ggpointdensity) library(ggsci) library(ggtext) library(patchwork) library(circlize) library(viridis) library(ggrepel) library(ComplexHeatmap) # visuals - format library(kableExtra) library(knitr) library(gridExtra) library(grid) # calculation library(matrixStats) library(umap) library(fpc) library(dbscan) library(multimode) ``` ```{r} #| label: load additional scripts #| message: false source("../styles.R") source("../helper.R") ``` # Prepartion of regions At first binding sites closer to each other than 55nt are merged. Merged regions are then symmetrically extended to form 81nt wide bins. ```{r} #| label: load clip data #| message: false load("/Users/mirko/Projects/sf3b1/02_markdowns/03_clean/01_bindingSites/data/bsTranscript.rda") # Load clip data clipFilesWt = "/Users/mirko/Projects/sf3b1/01_data_subsamp/wt/cov/replicate" clipFilesMut = "/Users/mirko/Projects/sf3b1/01_data_subsamp/mut/cov/replicate" clipFiles = c(clipFilesWt, clipFilesMut) clipFiles = list.files(clipFiles, pattern = ".bw$", full.names = TRUE) clipFilesP = clipFiles[grep(clipFiles, pattern = "Plus")] clipFilesM = clipFiles[grep(clipFiles, pattern = "Minus")] # Organize clip data in dataframe colData = data.frame( id = c(1:5), condition = factor(c("WT", "WT", "WT", "WT", "WT"), levels = c("WT")), clPlus = clipFilesP, clMinus = clipFilesM) # Make BindingSiteFinder object bds = BSFDataSetFromBigWig(ranges = bsTranscript, meta = colData) ``` ## Distance pattern for binding sites in introns ::: {.panel-tabset} ### Zoom-Out ```{r, fig.width=4, fig.height=4} #| message: false #| warning: false #| fig-width: 4 #| fig-height: 4 #| fig-cap: Distance from each binding site to the next closest neighbor. bsIntron = subset(bsTranscript, bsTranscript$region == "intron") dist = distanceToNearest(bsIntron) %>% as.data.frame() bsIntron$dist = dist$distance ggplot(dist, aes(x = log10(distance+1))) + geom_histogram(bins = 100, color = "black") + theme_nice() + labs( title = "Distance to nearest binding site", x = "Distance +1 (nt) [log10]", y = "Count") ``` ### Zoom-In ```{r, fig.width=4, fig.height=4} #| message: false #| warning: false #| fig-width: 4 #| fig-height: 4 #| fig-cap: Distance from each binding site to the next closest neighbor in a range of 50 nt. ggplot(dist, aes(x = distance)) + geom_histogram(binwidth = 1, color = "black") + xlim(-1,50) + theme_nice() + labs( title = "Distance to nearest binding site", x = "Distance (nt) [0-50]", y = "Count") + geom_vline(xintercept = 7, linetype = "dashed") ``` ::: ## Transform the iCLIP signal To avoid that the height of the coverage influences our downstream clustering, the signal in each bin is scaled between 0 and 1. Next a spline transformation is applied to produced a smoothed version of the coverage, which boosts classification performance. After testing different lambda values (0.2, 0.3, 0.4) we decided from the heatmaps below to continue with `lambda = 0.2` and a dimension `dim = 150`. ```{r} #| label: prepare ranges #| message: false # combine all BS within 41 nt range mergedRange = reduce(bsIntron, min.gapwidth = 55, with.revmap = TRUE) mergedRange$width = width(mergedRange) # add bs count info nBS = sapply(mergedRange$revmap, length) mcols(mergedRange)$nBS = nBS # resize ranges to their center position and extend all ranges to 81 nt rngSel = resize(granges(mergedRange), fix = "center", width = 81) names(rngSel) = 1:length(rngSel) export(mergedRange, "./data/mergedRange.bed", format = "BED") export(rngSel, "./data/rngSel.bed", format = "BED") saveRDS(mergedRange, file = "./data/mergedRange.rds") saveRDS(rngSel, file = "./data/rngSel.rds") ``` ```{r} #| label: prepare matrix #| message: false bdsSel = setRanges(bds, rngSel) cov = coverageOverRanges(bdsSel, returnOptions = "merge_all_replicates") masterMM = cov # normalize matrix normMM = minMaxNorm(masterMM) # # apply different smoothing levels smoothMM_par1 <- t(apply(normMM, 1, smoothing, lambda=0.2, dim=151)) smoothMM_par2 <- t(apply(normMM, 1, smoothing, lambda=0.3, dim=151)) smoothMM_par3 <- t(apply(normMM, 1, smoothing, lambda=0.4, dim=151)) saveRDS(smoothMM_par1, file = "./data/smoothMM_par1.rds") saveRDS(smoothMM_par2, file = "./data/smoothMM_par2.rds") saveRDS(smoothMM_par3, file = "./data/smoothMM_par3.rds") ``` ## Smooting levels ::: {.panel-tabset} ### Smooth 0.2 ```{r, fig.width=4, fig.height=5} #| message: false #| warning: false #| fig-width: 4 #| fig-height: 5 #| fig-cap: Heatmaps of smoothed binding site signal with level 0.2 set.seed(1234) custom.col = viridis(10, option = "G", direction = -1) idx = sample(1:(nrow(smoothMM_par1)), 500) m1 = smoothMM_par1[idx,] m2 = smoothMM_par2[idx,] m3 = smoothMM_par3[idx,] h1 = Heatmap(m1, column_title = "Smooth 0.2", name = "Xlinks", cluster_columns = F, cluster_rows = T, show_row_names = FALSE, show_column_names = T, border = T, col = custom.col, heatmap_legend_param = list( at = c(0,0.5, 1), labels = c("0", "0.5", "1"), legend_width = unit(6, "cm"), title_position = "topleft", direction = "horizontal" ), use_raster = TRUE ) draw(h1, heatmap_legend_side = "bottom") ``` ### Smooth 0.3 ```{r, fig.width=4, fig.height=5} #| message: false #| warning: false #| fig-width: 4 #| fig-height: 5 #| fig-cap: Heatmaps of smoothed binding site signal with level 0.3 h2 = Heatmap(m2, column_title = "Smooth 0.3", name = "Xlinks", cluster_columns = F, cluster_rows = T, show_row_names = FALSE, show_column_names = T, border = T, col = custom.col, heatmap_legend_param = list( at = c(0,0.5, 1), labels = c("0", "0.5", "1"), legend_width = unit(6, "cm"), title_position = "topleft", direction = "horizontal" ), use_raster = TRUE ) draw(h2, heatmap_legend_side = "bottom") ``` ### Smooth 0.4 ```{r, fig.width=4, fig.height=5} #| message: false #| warning: false #| fig-width: 4 #| fig-height: 5 #| fig-cap: Heatmaps of smoothed binding site signal with level 0.3 h3 = Heatmap(m3, column_title = "Smooth 0.4", name = "Xlinks", cluster_columns = F, cluster_rows = T, show_row_names = FALSE, show_column_names = T, border = T, col = custom.col, heatmap_legend_param = list( at = c(0,0.5, 1), labels = c("0", "0.5", "1"), legend_width = unit(6, "cm"), title_position = "topleft", direction = "horizontal" ), use_raster = TRUE ) draw(h3, heatmap_legend_side = "bottom") ``` ::: # UMAP classification of intronic binding regions UMAP classification is calculated with settings as proposed in the STOATY dive approach for coverage shape based binding site clustering. Setting details are: `n_epochs = 5000`, `n_components = 2`, `min_dist = 0.01` and `n_neighbors = 5`. ```{r} #| label: load umap #| message: false umapDf_par1 = readRDS("./data/umapDf_par1.rds") ``` # Density based clustering (DBSCAN) of UMAP result UMAP results are clustered with DBSCAN (density based clustering). This method defines cluster on the observed density, based on the definition of cluster centers. In particular the minimal number of points that are needed to form a cluster and the distance that points need to have to be assigned to a cluster. The following settings were used: `eps = 0.3` and `MinPts = 150`. ## DBSCAN results ::: {.panel-tabset} ### K Distances ```{r, fig.width=4, fig.height=4} #| message: false #| warning: false #| fig-width: 4 #| fig-height: 4 #| fig-cap: K nearest neighbors plot with k 150. K equals the MinPts option in the dbscan. Line at 0.3 represents eps param. set.seed(1234) dbscan::kNNdistplot(umapDf_par1$layout, k = 150) abline(h = 0.3, lty = 2) ``` ### Cleand clusters ```{r, fig.width=4, fig.height=4} #| message: false #| warning: false #| fig-width: 4 #| fig-height: 4 #| fig-cap: DBSCAN clustering of UMAP transformation for regions with smooting 0.2 and dbscan clustering; Outlier cluster 0 removed. grouping = dbscan::dbscan(umapDf_par1$layout, eps = 0.3, minPts = 150) grouping$cluster = ifelse(grouping$cluster == 2, 3, ifelse(grouping$cluster == 3, 2, grouping$cluster)) df1 = data.frame(x = umapDf_par1$layout[,1], y = umapDf_par1$layout[,2], group = factor(grouping$cluster)) df1 = subset(df1, group != 0) ggplot(df1, aes(x, y)) + ggrastr::rasterise(geom_pointdensity(size = 0.2), dpi = 300) + theme_pub() + scale_color_viridis(option = "A") + theme(legend.position = "top", aspect.ratio = 1) + ggforce::geom_mark_ellipse(aes(label = group, group = group), alpha = 0, label.fontsize = 12, expand = unit(2, "mm")) + guides(color = guide_colorbar(title.position = 'top', title.hjust = 0.5, barwidth = unit(20, 'lines'), barheight = unit(.5, 'lines'))) + scale_y_continuous(limits = c(-12, 20)) + scale_x_continuous(limits = c(-12, 8)) + labs( x = "UMAP 1", y = "UMAP 2", ) ``` ::: ## Cluster examples as heatmap ```{r,fig.width=6, fig.height=4} #| message: false #| warning: false #| fig-width: 6 #| fig-height: 4 #| fig-cap: moothed crosslink heatmaps split by kmeans clustering. Subsetted to top (head) 300. set.seed(1234) s = split.data.frame(smoothMM_par1, grouping$cluster) # s$rest = rbind(s$`1`, s$`2`, s$`3`) s = lapply(s, head, n = 300) custom.col = viridis(10, option = "G", direction = -1) clustRow = FALSE # annotate with top col sums profile m1 df1 = data.frame(sums = colMeans(s$`1`)) haMeans1 = HeatmapAnnotation(cov = anno_barplot(df1, gp = gpar(fill = "#595959", col = "#595959")), height = unit(2, "cm"), show_annotation_name = F) # annotate with top col sums profile m2 df2 = data.frame(sums = colMeans(s$`2`)) haMeans2 = HeatmapAnnotation(cov = anno_barplot(df2, gp = gpar(fill = "#595959", col = "#595959")), height = unit(2, "cm"), show_annotation_name = F) # annotate with top col sums profile m3 df3 = data.frame(sums = colMeans(s$`3`)) haMeans3 = HeatmapAnnotation(cov = anno_barplot(df3, gp = gpar(fill = "#595959", col = "#595959")), height = unit(2, "cm"), show_annotation_name = F) h1 = Heatmap(s$`1`, column_title = "Cluster 1", name = "Xlinks", cluster_rows = clustRow, cluster_columns = F, show_row_names = FALSE, show_column_names = T, border = T, col = custom.col, top_annotation = haMeans1, heatmap_legend_param = list( at = c(0,0.5, 1), labels = c("0", "0.5", "1"), legend_width = unit(6, "cm"), title_position = "topleft", direction = "horizontal" ), use_raster = TRUE ) h2 = Heatmap(s$`2`, column_title = "Cluster 2", name = "Xlinks", cluster_rows = clustRow, cluster_columns = F, show_row_names = FALSE, top_annotation = haMeans2, show_column_names = T, border = T, col = custom.col, use_raster = TRUE ) h3 = Heatmap(s$`3`, column_title = "Cluster 3", name = "Xlinks", cluster_rows = clustRow, cluster_columns = F, show_row_names = FALSE, top_annotation = haMeans3, show_column_names = T, border = T, col = custom.col, use_raster = TRUE ) l = h1 + h2 + h3 draw(l, heatmap_legend_side = "bottom") ``` ```{r} #| label: export clustering #| message: false mcols(rngSel)$MajorCluster = grouping$cluster rngExport = rngSel names(rngExport) = ifelse(rngExport$MajorCluster == 1, paste0("SinglePeak_", names(rngExport)), ifelse(rngExport$MajorCluster == 2, paste0("DoubleNarrow_", names(rngExport)), ifelse(rngExport$MajorCluster == 3, paste0("DoubleWide_", names(rngExport)), paste0("Rest_",names(rngExport))))) mcols(rngExport)$itemRgb = ifelse(rngExport$MajorCluster == 1, viridis(option = "G", n = 5)[2], ifelse(rngExport$MajorCluster == 2, viridis(option = "G", n = 5)[3], ifelse(rngExport$MajorCluster == 3, viridis(option = "G", n = 5)[4], "Grey"))) rtracklayer::export(rngExport, "./data/rngClassified.bed", format = "BED", trackLine = new("BasicTrackLine", name="Ranges major cluster", itemRgb = TRUE) ) ``` # Features of the clustered groups ## Explorative plots for result clusters. ```{r, fig.width=10, fig.height=10} #| message: false #| warning: false #| fig-width: 10 #| fig-height: 10 #| fig-cap: Number of binding sites in merged regions and clusters.} Binding site overlaps counted in merged ranges (sizes is variable here). s = split.data.frame(smoothMM_par1, grouping$cluster) df = data.frame(cluster = factor(names(s)), size = sapply(s, nrow)) p0 = ggplot(df, aes(x = cluster, y = size)) + geom_col() + theme_pub() + labs(y = "N") + scale_y_log10() + geom_text(aes(label = size), vjust = 1.3, color = "white") + labs( title = "Ranges per group", x = "Cluster", y = "Count" ) names(mergedRange) = 1:length(mergedRange) rngC1 = mergedRange[names(mergedRange) %in% rownames(s$`1`)] rngC2 = mergedRange[names(mergedRange) %in% rownames(s$`2`)] rngC3 = mergedRange[names(mergedRange) %in% rownames(s$`3`)] bsC1 = subsetByOverlaps(bsIntron, rngC1) bsC2 = subsetByOverlaps(bsIntron, rngC2) bsC3 = subsetByOverlaps(bsIntron, rngC3) df = data.frame(cluster = rep(c("C1", "C2", "C3"),2), type = c(rep("BS",3),rep("Region",3)), value = c(length(bsC1), length(bsC2), length(bsC3), length(rngC1), length(rngC2), length(rngC3)) ) p1 = ggplot(df, aes(x = cluster, y = value, fill = type)) + geom_col(position = "fill") + theme_pub() + theme(legend.position = "top") + scale_fill_npg() + labs( x = "Cluster", y = "Counts", fill = "Type" ) p2 = ggplot(df, aes(x = cluster, y = value, fill = type)) + geom_col(position = "dodge") + theme_pub() + theme(legend.position = "top") + scale_fill_npg() + labs( x = "Cluster", y = "Counts", fill = "Type" ) # names(mergedRange) = 1:length(mergedRange) # s = split.data.frame(smoothMM_par1, grouping$cluster) df1 = data.frame(N = mergedRange$nBS[names(mergedRange) %in% rownames(s$`1`)], cluster = "C1") df2 = data.frame(N = mergedRange$nBS[names(mergedRange) %in% rownames(s$`2`)], cluster = "C2") df3 = data.frame(N = mergedRange$nBS[names(mergedRange) %in% rownames(s$`3`)], cluster = "C3") df = rbind(df1,df2,df3) df$N[df$N > 7] = 7 df = table(df$N, df$cluster) %>% as.data.frame() p3 = ggplot(df, aes(x = Var1, y = Freq+1, fill = Var2)) + geom_col(position = "dodge") + scale_y_log10() + theme_pub() + scale_fill_npg() + theme(legend.position = "top") + labs( title = "Binding sites per group and cluster", x = "Number of Binding sites", y = "Count", fill = "Cluster" ) p4 = ggplot(df, aes(x = Var1, y = Freq+1, fill = Var2)) + geom_col(position = "fill") + # scale_y_log10() + theme_pub() + scale_fill_npg() + theme(legend.position = "top") + labs( title = "Binding sites per group and cluster", x = "Number of Binding sites", y = "Count", fill = "Cluster" ) p5 = ggplot(df, aes(x = Var1, y = Freq+1, fill = Var2)) + geom_col(position = "dodge") + scale_y_log10() + theme_pub() + scale_fill_npg() + theme(legend.position = "top") + labs( title = "Binding sites per group and cluster", x = "Number of Binding sites", y = "Count", fill = "Cluster" ) + facet_grid(~Var2) (p2 + p1) / (p0 + p3) / (p4 + p5) ``` ```{r, fig.width=10, fig.height=10} #| message: false #| warning: false #| fig-width: 10 #| fig-height: 10 #| fig-cap: Number of binding sites in merged regions and clusters. Binding site overlaps counted in 81nt clustered ranges. s = split.data.frame(smoothMM_par1, grouping$cluster) df = data.frame(cluster = factor(names(s)), size = sapply(s, nrow)) p0 = ggplot(df, aes(x = cluster, y = size)) + geom_col() + theme_pub() + labs(y = "N") + scale_y_log10() + geom_text(aes(label = size), vjust = 1.3, color = "white") + labs( title = "Ranges per group", x = "Cluster", y = "Count" ) rngC1 = rngSel[names(rngSel) %in% rownames(s$`1`)] rngC2 = rngSel[names(rngSel) %in% rownames(s$`2`)] rngC3 = rngSel[names(rngSel) %in% rownames(s$`3`)] bsC1 = subsetByOverlaps(bsIntron, rngC1) bsC2 = subsetByOverlaps(bsIntron, rngC2) bsC3 = subsetByOverlaps(bsIntron, rngC3) df = data.frame(cluster = rep(c("C1", "C2", "C3"),2), type = c(rep("BS",3),rep("Region",3)), value = c(length(bsC1), length(bsC2), length(bsC3), length(rngC1), length(rngC2), length(rngC3)) ) p1 = ggplot(df, aes(x = cluster, y = value, fill = type)) + geom_col(position = "fill") + theme_pub() + theme(legend.position = "top") + scale_fill_npg() + labs( x = "Cluster", y = "Counts", fill = "Type" ) p2 = ggplot(df, aes(x = cluster, y = value, fill = type)) + geom_col(position = "dodge") + theme_pub() + theme(legend.position = "top") + scale_fill_npg() + labs( x = "Cluster", y = "Counts", fill = "Type" ) df1 = data.frame(N = countOverlaps(rngC1, bsIntron), cluster = "C1") df2 = data.frame(N = countOverlaps(rngC2, bsIntron), cluster = "C2") df3 = data.frame(N = countOverlaps(rngC3, bsIntron), cluster = "C3") df = rbind(df1,df2,df3) df$N[df$N > 5] = 5 df = table(df$N, df$cluster) %>% as.data.frame() p3 = ggplot(df, aes(x = Var1, y = Freq+1, fill = Var2)) + geom_col(position = "dodge") + scale_y_log10() + theme_pub() + scale_fill_npg() + theme(legend.position = "top") + labs( title = "Binding sites per group and cluster", x = "Number of Binding sites", y = "Count", fill = "Cluster" ) + geom_text(aes(label = Freq), vjust = 0.5, hjust = 1.2, position = position_dodge(width = .9), angle = 90) p4 = ggplot(df, aes(x = Var1, y = Freq+1, fill = Var2)) + geom_col(position = "fill") + # scale_y_log10() + theme_pub() + scale_fill_npg() + theme(legend.position = "top") + labs( title = "Binding sites per group and cluster", x = "Number of Binding sites", y = "Count", fill = "Cluster" ) + scale_x_discrete(breaks=c("1", "2", "3", "4", "5"), labels=c("1", "2", "3", "4", "5+")) p5 = ggplot(df, aes(x = Var1, y = Freq+1, fill = Var2)) + geom_col(position = "dodge") + scale_y_log10() + theme_pub() + scale_fill_npg() + theme(legend.position = "top") + labs( title = "Binding sites per group and cluster", x = "Number of Binding sites", y = "Count", fill = "Cluster" ) + facet_grid(~Var2) (p2 + p1) / (p0 + p3) / (p4 + p5) ``` ## Binding sites per cluster selection ::: {.panel-tabset} ### V1 ```{r, fig.width=4, fig.height=4} #| message: false #| warning: false #| fig-width: 4 #| fig-height: 4 #| fig-cap: Binding sites per cluster. df2 = df df2$Var1 = ifelse(df2$Var1 == 5, "5+", df$Var1) p1 = ggplot(df2, aes(x = Var2, y = Freq+1, fill = Var1)) + geom_col(position = "fill") + theme_pub() + scale_fill_npg() + theme(legend.position = "top") + labs( title = "Number of binding sites per cluster", x = "Cluster", y = "Fraction", fill = "Binding sites" ) + theme(aspect.ratio = 1) p1 ``` ### V2 ```{r, fig.width=4, fig.height=4} #| message: false #| warning: false #| fig-width: 4 #| fig-height: 4 #| fig-cap: Binding sites per cluster. p2 = ggplot(df2, aes(x = Var2, y = Freq+1, fill = Var1)) + geom_col(position = "fill") + theme_pub() + scale_fill_grey() + theme(legend.position = "top") + labs( title = "Number of binding sites per cluster", x = "Cluster", y = "Fraction", fill = "Binding sites" ) + theme(aspect.ratio = 1) p2 ``` ### V3 ```{r, fig.width=4, fig.height=4} #| message: false #| warning: false #| fig-width: 4 #| fig-height: 4 #| fig-cap: Binding sites per cluster. p3 = ggplot(df2, aes(x = Var2, y = Freq+1, fill = Var1)) + geom_col(position = "fill") + theme_pub() + scale_fill_brewer(palette = 1, direction = 1) + theme(legend.position = "top") + labs( title = "Number of binding sites per cluster", x = "Cluster", y = "Fraction", fill = "Binding sites" ) + theme(aspect.ratio = 1) p3 ``` ::: ```{r} #| label: make table #| message: false df1 = data.frame(N = countOverlaps(rngC1, bsIntron), cluster = "C1") df2 = data.frame(N = countOverlaps(rngC2, bsIntron), cluster = "C2") df3 = data.frame(N = countOverlaps(rngC3, bsIntron), cluster = "C3") df = rbind(df1,df2,df3) d = df %>% group_by(cluster) %>% summarize(mean = mean(N), median = median(N)) %>% as.data.frame() kable(d, caption = "Number of BS per cluster") %>% kable_styling("striped") %>% scroll_box(width = "100%") ``` ## Distance to splice site per cluster ```{r} #| label: load gene annotation #| message: false load("/Users/mirko/Projects/Annotations/human/gencode_36/filtered/gencode_v36_filtered.rda") anno.db = loadDb("/Users/mirko/Projects/Annotations/human/gencode_36/filtered/gencode_v36_filtered.sqlite") gns = genes(anno.db) idx = match(gns$gene_id, anno$gene_id) elementMetadata(gns) = cbind(elementMetadata(gns), elementMetadata(anno)[idx,]) ``` ```{r} #| label: calculate distances #| message: false x0 = subset(anno, transcript_type == "protein_coding") x1 = subset(x0, type == "exon" | type == "CDS") x2 = subset(x1, tag == "exp_conf" | tag == "CCDS" | tag == "basic") exn = exons(anno.db) exn = subsetByOverlaps(exn, x2, type = "within") export(granges(exn), "./data/exn.bed", format = "BED") spliceSites3 = unique(resize(exn, fix = "start", width = 1)) spliceSites5 = unique(resize(exn, fix = "end", width = 1)) calcDist <- function(rng) { # identify the next downstream 3'SS for each given range idx3 = precede(rng, spliceSites3) # identify the next downstream 3'SS for each given range idx5 = follow(rng, spliceSites5) if (any(is.na(idx3), is.na(idx5))) { removeIdx3 = which(is.na(idx3)) removeIdx5 = which(is.na(idx5)) removeIdx = c(removeIdx3, removeIdx5) idx3 = idx3[-removeIdx] idx5 = idx5[-removeIdx] rng = rng[-removeIdx] } # calculate distance between given range and associated splice site d1 = distance(rng, spliceSites3[idx3]) d2 = distance(rng, spliceSites5[idx5]) df = data.frame(d3ss = d1, d5ss = d2) return(df) } ``` ::: {.panel-tabset} ### Distance ```{r, fig.width=4, fig.height=4} #| message: false #| warning: false #| fig-width: 4 #| fig-height: 4 #| fig-cap: Distances to splice sites. Distance is calculated from the outer edge of the outer binding sites for each region to the respective 3'ss and 5'ss splice site" s = split.data.frame(smoothMM_par1, grouping$cluster) originalRange = mergedRange originalRange = granges(originalRange) names(originalRange) = names(rngSel) rngC1 = originalRange[names(originalRange) %in% rownames(s$`1`)] rngC2 = originalRange[names(originalRange) %in% rownames(s$`2`)] rngC3 = originalRange[names(originalRange) %in% rownames(s$`3`)] # calculate distance df1 = calcDist(rngC1) %>% as.data.frame() %>% mutate(clust = "C1: single") %>% pivot_longer(-clust) df2 = calcDist(rngC2) %>% as.data.frame() %>% mutate(clust = "C2: double-narrow") %>% pivot_longer(-clust) df3 = calcDist(rngC3) %>% as.data.frame() %>% mutate(clust = "C3: double-wide") %>% pivot_longer(-clust) df = rbind(df1,df2,df3) df$value[df$value == 0] = 1 p1 = ggplot(df, aes(x = clust, y = log10(value), fill = name)) + geom_boxplot(outlier.shape = NA) + theme_pub() + scale_fill_npg() + labs( x = "Cluster", y = "Distance to splice site (log10)", fill = "Splice site" ) p1 ``` ### Distance ratio ```{r, fig.width=4, fig.height=4} #| message: false #| warning: false #| fig-width: 4 #| fig-height: 4 #| fig-cap: Distances to splice sites. Distance is calculated from the outer edge of the outer binding sites for each region to the respective 3'ss and 5'ss splice site" df1 = calcDist(rngC1) %>% as.data.frame() %>% mutate(clust = "C1: single", r = d3ss/ d5ss) df2 = calcDist(rngC2) %>% as.data.frame() %>% mutate(clust = "C2: double-narrow", r = d3ss/ d5ss) df3 = calcDist(rngC3) %>% as.data.frame() %>% mutate(clust = "C3: double-wide", r = d3ss/ d5ss) df = rbind(df1,df2,df3) p2 = ggplot(df, aes(x = clust, y = log2(r), fill = clust)) + geom_boxplot(outlier.shape = NA) + theme_pub() + scale_fill_manual(values = viridis(n = 10, option = "mako")[c(3,5,7)]) + theme(legend.position = "none") + labs( x = "Cluster", y = "Splice site distance ratio (log2(3'ss/5'ss))" ) p2 ``` ::: # UMAP subclassification of major cluster 3 - double-wide pattern To further refine the positioning of double-peaks within the cluster 3 groups, we performed a second round of UMAP + DBSCAN. After cleaning resulting clusters each cluster gets assigned to a group based on the distance of the modes between both double-peaks. ```{r} #| label: smooth subcluster 3 #| message: false selMM_g3 = masterMM[grouping$cluster == 3,] normMM_g3 = apply(selMM_g3, 1, function(x){ n = ((x - min(x)) / (max(x) - min(x))) return(n) }) normMM_g3 = t(normMM_g3) smoothMM_g3_par1 <- t(apply(normMM_g3, 1, smoothing, lambda=0.1, dim=500)) saveRDS(smoothMM_g3_par1, "./data/smoothMM_g3_par1.rds") ``` ```{r} #| label: load umap for subcluster 3 #| message: false umapDf_g3_par1 = readRDS("./data/umapDf_g3_par1.rds") ``` ## DBSCAN clustering results ::: {.panel-tabset} ### K Distances ```{r, fig.width=4, fig.height=4} #| message: false #| warning: false #| fig-width: 4 #| fig-height: 4 #| fig-cap: K nearest neighbors plot with k 150. K equals the MinPts option in the dbscan. Line at 0.3 represents eps param. set.seed(1234) dbscan::kNNdistplot(umapDf_g3_par1$layout, k = 60) abline(h = 0.22, lty = 2) ``` ### Cleand clusters ```{r, fig.width=4, fig.height=4} #| message: false #| warning: false #| fig-width: 4 #| fig-height: 4 #| fig-cap: DBSCAN clustering of UMAP transformation for regions with smooting 0.2 and dbscan clustering; Outlier cluster 0 removed. groupingG3 = dbscan::dbscan(umapDf_g3_par1$layout, eps = 0.23, MinPts = 60) df1 = data.frame(x = umapDf_g3_par1$layout[,1], y = umapDf_g3_par1$layout[,2], group = factor(groupingG3$cluster)) df1 = df1 %>% subset(group != "0") p3 = ggplot(df1, aes(x, y, color = group)) + # geom_point(size = 0.2) + ggrastr::geom_point_rast(size = 0.2) + theme_pub() + # scale_color_manual(values = viridis(n = 6, option = "mako")[2:8]) + guides(colour = guide_legend(override.aes = list(size=2), nrow = 4)) + theme(legend.position = "none") + labs(title = "Smooth 0.1", color = "C") p3 ``` ::: ## Order cluster groups by peak mode distance ::: {.panel-tabset} ### Scatter ```{r, fig.width=4, fig.height=4} #| message: false #| warning: false #| fig-width: 4 #| fig-height: 4 #| fig-cap: Sub-cluster 3 groups matched by double peak mode distances set.seed(1234) # split smoothed coverage matrix by clustering groups s = split.data.frame(smoothMM_g3_par1, groupingG3$cluster) tmp = lapply(1:length(s), function(x){ z = floor(ncol(s[[1]]) / 2) d = data.frame(Group = factor(rep(x, ncol(s[[x]]))), value = colMeans(s[[x]], na.rm = TRUE), pos = -246:247) }) df = dplyr::bind_rows(tmp, .id = "variable") df$Group = as.factor(as.numeric(df$Group)-1) # calc new position dimOrg = 81 dimSmooth = 500 dimSf = dimSmooth / dimOrg df$newPos = df$pos / dimSf # split by group df = df[df$Group != 0,] dfSplit = split(df, df$Group) dfSplit = lapply(dfSplit, function(x){ data.frame(v = rep(x$newPos, ifelse(x$value < 0, x$value*(-1000), x$value*1000))) }) dfSplit = dplyr::bind_rows(dfSplit, .id = "variable") dfSplit = dfSplit[dfSplit$variable != 0,] # calculate modes modeDf = split.data.frame(dfSplit, dfSplit$variable) modes = lapply(modeDf, function(x){ m = locmodes(x$v, mod0 = 2) d = m$locations return(d) }) modes = dplyr::bind_rows(modes, .id = "variable") %>% t() %>% as.data.frame() %>% tibble::rownames_to_column("cluster") %>% rename(mode1 = V1, mode2 = V3, antimode = V2) dfLine = modes %>% select(cluster, mode1, mode2) %>% pivot_longer(-cluster) %>% rename(variable = cluster) dist = length(modes$mode1:modes$mode2) dfDist = modes %>% group_by(cluster) %>% mutate(dist = length(mode1:mode2)) %>% select(cluster, dist) %>% rename(variable = cluster) fOrder = dfDist$variable[order(dfDist$dist)] dfSplit$variable = factor(dfSplit$variable, levels = fOrder) dfLine$variable = factor(dfLine$variable, levels = fOrder) dfDist$variable = factor(dfDist$variable, levels = fOrder) dfOrder = dfDist[order(dfDist$dist),] dfOrder$NewCluster = 1:nrow(dfOrder) idx = match(df1$group, dfOrder$variable) df1$NewCluster = as.factor(dfOrder$NewCluster[idx]) ggplot(subset(df1, NewCluster != 0), aes(x, y, fill = NewCluster)) + rasterize(geom_point(shape = 21, stroke = 0.3), dpi = 300) + theme_pub() + scale_fill_viridis(option = "turbo", discrete = T, direction = -1) + theme(legend.position = "none", aspect.ratio = 1) + scale_y_continuous(limits = c(-10, 10)) + scale_x_continuous(limits = c(-10, 15)) + labs( x = "UMAP 1", y = "UMAP 2", ) + ggforce::geom_mark_ellipse(aes(label = NewCluster, group = NewCluster), alpha = 0, label.fontsize = 12, label.buffer = unit(5, 'mm'), expand = unit(2, "mm"), label.minwidth = unit(5, "mm")) ``` ### Coverage ```{r, fig.width=4, fig.height=4} #| message: false #| warning: false #| fig-width: 4 #| fig-height: 4 #| fig-cap: Sub-cluster 3 smoothed crosslink densities split by density based clustering.} Sorted by double-mode distances. df = dfSplit dfOrder = dfDist[order(dfDist$dist),] dfOrder$NewCluster = 1:nrow(dfOrder) idx = match(df$variable, dfOrder$variable) df$NewCluster = dfOrder$NewCluster[idx] df$name = paste0("cluster: ", df$NewCluster, ", distance: ", dfOrder$dist[idx]) df$name = factor(df$name) df$name = factor(df$name, levels = as.character(unique(df$name)[as.numeric(levels(df$variable))])) p = ggplot(df, aes(x = v, y = name, fill = ..density..)) + rasterise(geom_density_ridges_gradient(scale = 10, bandwidth = 0.1, color = "black", size = 0.5),dpi = 300) + # theme_pub() + theme_pub() + theme(legend.position = "top") + scale_y_discrete(expand = expand_scale(mult = c(0, 0.15))) + scale_fill_gradient2(position="top" , low = "white", high = "#366A9FFF", mid = "348AA6FF", midpoint = 0.05, breaks = c(0.01, 0.05, 0.1)) + labs( title = "Sub-cluster by double-mode distance", x = "Relative distance (nt)", y = "Cluster" ) p ``` ::: # Session Information ```{r, session_info, include=TRUE, echo=TRUE, results='markup'} sessionInfo() ```