-
Notifications
You must be signed in to change notification settings - Fork 2
/
flowsom.Rmd
262 lines (220 loc) · 10.3 KB
/
flowsom.Rmd
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
---
title: "Clustering 1.3M cells with FlowSOM"
author: ""
output: html_document
---
```{r}
suppressPackageStartupMessages(library(TENxBrainData))
suppressPackageStartupMessages(library(scater))
suppressPackageStartupMessages(library(flowCore))
suppressPackageStartupMessages(library(FlowSOM))
suppressPackageStartupMessages(library(scran))
suppressPackageStartupMessages(library(pheatmap))
suppressPackageStartupMessages(library(tibble))
suppressPackageStartupMessages(library(dplyr))
suppressPackageStartupMessages(library(reshape2))
suppressPackageStartupMessages(library(ggplot2))
suppressPackageStartupMessages(library(Rtsne))
suppressPackageStartupMessages(library(mclust))
```
## Read data
```{r}
tenx <- TENxBrainData()
sce <- readRDS("objects/sce.rds")
tenx <- tenx[, colnames(sce)] # drop 19,672 cells from the raw TENxBrainData
counts(sce) <- counts(tenx) # overwrite inbuilt absolute path
system.time(sce <- scater::normalize(sce)) # generate normalized expression values
```
## Read rSVD results
```{r}
svd <- readRDS("objects/rsvd.rds")
pca <- sweep(svd$u, 2, svd$d, "*")
colnames(pca) <- paste0("PC", seq_len(ncol(pca)))
dim(pca)
```
## Apply FlowSOM to cluster the cells based on the first 50 PCs
```{r}
set.seed(123)
system.time(ff <- flowFrame(exprs = pca))
system.time(fSOM <- FlowSOM::ReadInput(ff, compensate = FALSE, transform = FALSE,
scale = FALSE, silent = TRUE))
system.time(fSOM <- FlowSOM::BuildSOM(fSOM, silent = TRUE, xdim = 13, ydim = 13))
system.time(metaClustering <- metaClustering_consensus(fSOM$map$codes, k = 16))
```
## Add cluster info to data object
```{r}
colData(sce)$som100 <- fSOM$map$mapping[, 1]
colData(sce)$sommeta <- metaClustering[fSOM$map$mapping[, 1]]
```
## Add rownames
```{r}
rownames(sce) <- paste0(rowData(sce)$Ensembl, ".", rowData(sce)$Symbol)
```
## Plot PCA representation
```{r pca, fig.width = 8, fig.height = 8}
dim(sce)
dim(pca)
cols <- c("#DC050C", "#E8601C", "#7BAFDE", "#1965B0", "#B17BA6",
"#882E72", "#F1932D", "#F6C141", "#F7EE55", "#4EB265",
"#90C987", "#CAEDAB", "#777777", "black", "cyan", "pink")
names(cols) <- as.character(seq_len(length(cols)))
dfpca <- data.frame(pca, som100 = factor(colData(sce)$som100),
sommeta = factor(colData(sce)$sommeta),
library_id = factor(colData(sce)$Library),
mouse = factor(colData(sce)$Mouse),
stringsAsFactors = FALSE)
print(ggplot(dfpca, aes(x = PC1, y = PC2, color = sommeta)) +
geom_point(size = 0.5) + scale_color_manual(values = cols) +
theme_bw())
print(ggplot(dfpca, aes(x = PC3, y = PC4, color = sommeta)) +
geom_point(size = 0.5) + scale_color_manual(values = cols) +
theme_bw())
print(ggplot(dfpca, aes(x = PC5, y = PC6, color = sommeta)) +
geom_point(size = 0.5) + scale_color_manual(values = cols) +
theme_bw())
print(ggplot(dfpca, aes(x = PC7, y = PC8, color = sommeta)) +
geom_point(size = 0.5) + scale_color_manual(values = cols) +
theme_bw())
print(ggplot(dfpca, aes(x = PC9, y = PC10, color = sommeta)) +
geom_point(size = 0.5) + scale_color_manual(values = cols) +
theme_bw())
```
Also plot PCA colored by library ID/mouse
```{r pca2}
print(ggplot(dfpca, aes(x = PC1, y = PC2, color = library_id)) +
geom_point(size = 0.5) + theme_bw())
print(ggplot(dfpca, aes(x = PC1, y = PC2, color = mouse)) +
geom_point(size = 0.5) + theme_bw())
```
## Find marker genes
```{r scran_markers}
system.time(scran_markers_all <- scran::findMarkers(sce,
clusters = colData(sce)$sommeta,
block = NULL,
design = NULL,
direction = "up",
pval.type = "any",
assay.type = "logcounts",
get.spikes = FALSE,
log.p = TRUE,
lfc = 0.5))
for (i in seq_len(length(scran_markers_all))) {
print(head(as.data.frame(scran_markers_all[[i]])))
}
scran_markers <- unique(unlist(lapply(scran_markers_all, function(w) {
rownames(subset(w, Top <= 1))
})))
scescranmarker <- sce[which(rownames(sce) %in% scran_markers), ]
logcounts_scranmarkers <- logcounts(scescranmarker)
rownames(logcounts_scranmarkers) <- rowData(scescranmarker)$Symbol
dfscran <- as.data.frame(logcounts_scranmarkers) %>% tibble::rownames_to_column("gene") %>%
reshape2::melt() %>% dplyr::left_join(as.data.frame(colData(scescranmarker)) %>%
tibble::rownames_to_column("variable") %>%
dplyr::select(variable, som100, sommeta)) %>%
dplyr::mutate(sommeta = factor(sommeta)) %>%
dplyr::mutate(som100 = factor(som100))
```
Heatmap of inferred marker genes
```{r scran_markers_heatmap, fig.height = 12, fig.width = 10}
dfsumscran <- dfscran %>% dplyr::group_by(gene, som100) %>%
dplyr::summarize(value = quantile(value, probs = 0.75)) %>%
tidyr::spread(som100, value) %>% as.data.frame()
rownames(dfsumscran) <- dfsumscran$gene
dfsumscran$gene <- NULL
dfsumscran <- dfsumscran[apply(dfsumscran, 1, sd) > 0, ]
pheatmap::pheatmap(dfsumscran, scale = "row", cluster_rows = TRUE, cluster_cols = TRUE,
annotation_col = data.frame(metaClust = factor(metaClustering), row.names = as.character(seq_len(length(metaClustering)))),
show_colnames = FALSE, show_rownames = TRUE, fontsize_row = 6,
annotation_colors = list(metaClust = cols))
```
## Look at known marker genes.
Marker genes for cell types in mouse brain were obtained from two recent publications:
- [http://science.sciencemag.org/content/347/6226/1138.full](http://science.sciencemag.org/content/347/6226/1138.full)
- [https://www.nature.com/articles/nn.4216](https://www.nature.com/articles/nn.4216)
```{r markers, fig.width = 24, fig.height = 24}
marker_genes <- read.csv("resources/marker_genes.csv", header = TRUE, as.is = TRUE) %>%
dplyr::arrange(population) %>% dplyr::select(gene, population) %>%
dplyr::distinct()
keep <- which(rowData(sce)$Symbol %in% unique(marker_genes$gene))
scemarker <- sce[keep, ]
logcounts_markers <- assays(scemarker)[["logcounts"]]
rownames(logcounts_markers) <- rowData(scemarker)$Symbol
df <- as.data.frame(logcounts_markers) %>% tibble::rownames_to_column("gene") %>%
reshape2::melt() %>% dplyr::left_join(as.data.frame(colData(scemarker)) %>%
tibble::rownames_to_column("variable") %>%
dplyr::select(variable, som100, sommeta)) %>%
dplyr::mutate(sommeta = factor(sommeta)) %>%
dplyr::mutate(som100 = factor(som100)) %>%
dplyr::left_join(marker_genes %>% dplyr::select(gene, population)) %>%
dplyr::mutate(genepop = paste0(gene, " (", population, ")")) %>%
dplyr::arrange(population) %>%
dplyr::mutate(gene = factor(gene, levels = unique(marker_genes$gene)))
print(ggplot(df, aes(x = sommeta, y = value, fill = sommeta)) +
geom_boxplot(outlier.size = 0.5) + facet_wrap(~ genepop, scales = "free_y") +
scale_fill_manual(values = cols) + theme_bw())
```
Check the FDRs of the marker genes in each of the clusters
```{r known_markers_fdr, fig.height = 12}
L <- do.call(rbind, lapply(structure(marker_genes$gene, names = marker_genes$gene),
function(g) {
do.call(rbind, lapply(seq_along(scran_markers_all), function(i) {
j <- grep(paste0("\\.", g, "$"), rownames(scran_markers_all[[i]]))
if (length(j) == 1) data.frame(gene = g, cluster = i,
log.FDR = scran_markers_all[[i]][j, "log.FDR"],
stringsAsFactors = FALSE)
else data.frame(gene = g, cluster = i, log.FDR = NA, stringsAsFactors = FALSE)
}))
}))
L <- tidyr::spread(L, cluster, log.FDR)
rownames(L) <- L$gene
L$gene <- NULL
L <- L[rowSums(is.na(L)) == 0, ]
pheatmap::pheatmap(exp(L), scale = "none", cluster_rows = TRUE, cluster_cols = TRUE)
```
Plot the individual marker gene expression in each of the 100 original FlowSOM
clusters, color by the final cluster assignment.
```{r markers-indiv, fig.width = 18}
for (g in unique(df$genepop)) {
print(ggplot(df %>% dplyr::filter(genepop == g),
aes(x = som100, y = value, fill = sommeta)) +
geom_boxplot(outlier.size = 0.5) + ggtitle(g) + theme_bw() +
theme(axis.text.x = element_text(angle = 90, hjust = 1, vjust = 0.5)) +
scale_fill_manual(values = cols))
}
```
## Heatmap of marker genes
We make a heatmap of the marker genes, summarized for the cells in each of the
original FlowSOM clusters. As a summarization function, we use the third
quartile.
```{r, fig.width = 10, fig.height = 12}
dfsum <- df %>% dplyr::group_by(genepop, som100) %>%
dplyr::summarize(value = quantile(value, probs = 0.75)) %>%
tidyr::spread(som100, value) %>% as.data.frame()
rownames(dfsum) <- dfsum$genepop
dfsum$genepop <- NULL
dfsum <- dfsum[apply(dfsum, 1, sd) > 0, ]
pheatmap::pheatmap(dfsum, scale = "row", cluster_rows = TRUE, cluster_cols = TRUE,
annotation_col = data.frame(metaClust = factor(metaClustering), row.names = as.character(seq_len(length(metaClustering)))),
show_colnames = FALSE, show_rownames = TRUE, fontsize_row = 6,
annotation_colors = list(metaClust = cols))
```
## t-SNE of subsampled data
Here we apply t-SNE to 10,000 randomly selected cells, using the 50 first PCs as
the input. We color the resulting representation by the assigned cluster.
```{r tsnesub}
set.seed(123)
subs <- sample(seq_len(nrow(pca)), 10000, replace = FALSE)
pcasub <- pca[subs, ]
rtsne_out <- Rtsne(as.matrix(pcasub), pca = FALSE, verbose = TRUE, perplexity = 30)
rtsne_out <- data.frame(rtsne_out$Y)
colnames(rtsne_out) <- c("tSNE1", "tSNE2")
rtsne_out$sommeta <- factor(colData(sce)$sommeta[subs])
print(ggplot(rtsne_out, aes(x = tSNE1, y = tSNE2, color = sommeta)) +
geom_point(size = 0.75) + scale_color_manual(values = cols) +
theme_bw())
```
## Session info
```{r}
date()
sessionInfo()
```