-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy path03b-braf-netgsa-bootstrap.R
365 lines (302 loc) · 12.2 KB
/
03b-braf-netgsa-bootstrap.R
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
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
## Henry Linder 2020 [email protected]
data.path <- "~/local/data/TCGA/formatted"
out.path <- "bootstrap"
library(magrittr)
library(dplyr)
library(readr)
library(devtools)
library(RColorBrewer)
library(foreach)
library(doMC)
registerDoMC(4)
library(ppcor)
library(netgsa)
source("color.R")
innames <- load("A-braf-expr-mirna.Rdata")
## Names of omics features
all.names <- dimnames(A)[[1]]
## Names of miRNA
mirna.A <- all.names[grepl("^hsa-", all.names)]
## Identify miRNA which have both 3p and 5p, keep only one, and remove
## its identifier as being "-3p" or "-5p": TGCA does not include this
## information
rm.col <- gsub("-[35]p", "", mirna.A) %>% duplicated %>% which
rm.col <- which(colnames(A) %in% mirna.A[rm.col])
A <- A[-rm.col,-rm.col]
rownames(A) <- gsub("-[35]p", "", rownames(A))
colnames(A) <- gsub("-[35]p", "", colnames(A))
## Gene and miRNA symbols
all.names <- dimnames(A)[[1]]
mirna.A <- all.names[grepl("^hsa-", all.names)]
genes.A <- all.names[!all.names %in% mirna.A]
## All miRNAs in TCGA data
mirna.tcga <- read_csv("all-mirna.csv") %>% use_series(miRNA)
## Convert mirDIP identifiers to their corresponding TCGA format
mirdip.to.tcga <- function(x) {
x %>%
gsub("^hsa-", "", .) %>%
gsub("^mir-", "MIR", .) %>%
gsub("^let-", "LET", .) %>%
toupper
}
## Get the TCGA labels for the mirDIP miRNA tags
mirna.tcga.as.des <- mirdip.to.tcga(mirna.A)
if (!all(mirna.A %in% mirna.tcga)) {
stop("Something got mismatched!")
}
## All feature names
feature.names <- c(genes.A, mirna.tcga.as.des)
infiles <- list.files(data.path, full.names=TRUE)
nfiles <- length(infiles)
source("functions.R")
all.cancers <- foreach(ix.cancer=1:length(infiles)) %dopar% {
## for (ix.cancer in 1:length(infiles)) {
infile <- infiles[ix.cancer]
cancer <- infile %>% basename %>% strsplit("-") %>% sapply(head, 1)
print(sprintf("[%d / %d] %s", ix.cancer, nfiles, cancer))
innames <- load(infile)
## miRNA tags
mirna.des <- des %>% filter(Platform == "miRNAExp") %>% use_series(GeneSymbol)
if (!all(mirna.tcga.as.des %in% mirna.des)) print(sprintf("Missing features in %s", cancer))
## Data is a miRNA or a gene we are interested in
des <- des %>% filter(GeneSymbol %in% feature.names)
all.features <- list()
## Check which data platforms are available for each feature
for (fn in feature.names) {
d <- des %>% filter(GeneSymbol == fn)
dd <- data.frame(GeneSymbol=fn,
Expression="geneExp" %in% d$Platform,
Copy.Number="copyNumber" %in% d$Platform,
Methylation="methylation" %in% d$Platform,
mi="miRNAExp" %in% d$Platform,
stringsAsFactors=FALSE)
all.features[[length(all.features)+1]] <- dd
}
all.features <- bind_rows(all.features)
## Get the data, name the rows
dat <- dat[des$ix,]
dat <- dat[,!apply(dat, 2, . %>% is.na %>% any)]
dat <- dat[,grepl("-[01]1", colnames(dat))]
## Remove missing values
rm.rows <- dat %>% apply(1, . %>% is.na %>% any) %>% which
if (length(rm.rows) > 0) {
rm.counts <- dat %>% apply(1, . %>% is.na %>% sum)
rm.counts <- rm.counts[rm.rows]
dat <- dat[-rm.rows,]
##
rm.des <- des %>% filter(tag %in% names(rm.rows))
des <- des %>% filter(!tag %in% rm.des$tag)
}
## Remove those features with too few unique values in each population
nunique <- apply(dat[,grepl("-01$", colnames(dat)), drop=FALSE], 1, . %>% unique %>% length)
rm.rows1 <- which(nunique < 5)
nunique <- apply(dat[,grepl("-11$", colnames(dat)), drop=FALSE], 1, . %>% unique %>% length)
rm.rows2 <- which(nunique < 5)
rm.rows <- c(rm.rows1, rm.rows2) %>% unique
if (length(rm.rows) > 0) {
dat <- dat[-rm.rows,]
keep.des <- des %>% filter(tag %in% rownames(dat))
des <- des %>% filter(tag %in% keep.des$tag)
}
## Rename the dimensions of A to include information on genomic
## platform
mirna.tcga.labels <- mirdip.to.tcga(mirna.A)
feature.labels <- c(paste0(genes.A, "-GE"),
paste0(mirna.tcga.labels, "-mi"))
A.cancer <- A
dimnames(A.cancer) <- list(feature.labels, feature.labels)
## number of miRNA per gene
## A.cancer[grepl("-GE$", rownames(A.cancer)),grepl("-mi$", colnames(A.cancer))] %>% rowSums
## AKT1-GE BRAF-GE MAP2K1-GE MAP2K2-GE MAPK1-GE MTOR-GE NRAS-GE PIK3CA-GE
## 14 0 40 0 88 15 61 26
## PTEN-GE RAF1-GE
## 130 16
##
## number of genes targeted by each miRNA
## A.cancer[grepl("-GE$", rownames(A.cancer)),grepl("-mi$", colnames(A.cancer))] %>% colSums %>% table
## 1 2 3 4 5
## 108 56 31 18 1
## g=214 total miRNAs
## Expand A to include miRNA, EMC features
n.genes <- length(genes.A)
n.mirna <- length(mirna.A)
emc <- rbind(diag(n.genes),
matrix(0, n.mirna, n.genes))
emc.cn <- emc
colnames(emc.cn) <- paste0(genes.A, "-CN")
emc.me <- emc
colnames(emc.me) <- paste0(genes.A, "-ME")
A1 <- cbind(A.cancer, emc.cn, emc.me)
A2 <- matrix(0,
nrow=2*n.genes,
ncol=ncol(A) + 2*n.genes,
dimnames=list(c(colnames(emc.cn), colnames(emc.me)),
colnames(A1)))
A.cancer <- rbind(A1, A2)
ix.mir <- grepl("-mi$", colnames(A.cancer)) %>% which
ix.cn <- grepl("-CN$", colnames(A.cancer)) %>% which
ix.me <- grepl("-ME$", colnames(A.cancer)) %>% which
ix.ge <- grepl("-GE$", colnames(A.cancer)) %>% which
ix <- c(ix.mir, ix.cn, ix.me, ix.ge)
A.cancer <- A.cancer[ix, ix]
## Features we removed because of missing values or insufficiently
## many unique values
feat.A <- rownames(A.cancer)
feat.dat <- rownames(dat)
rm.feat <- which(!feat.A %in% feat.dat)
A.cancer <- A.cancer[-rm.feat,-rm.feat]
if (!all(sprintf("%s-GE", genes.A) %in% rownames(A.cancer))) {
print("Too many missing values")
return()
}
x <- dat[rownames(A.cancer),]
if (ncol(x) == 0) {
print("no subjects")
return()
}
keep.subj <- grepl("-[01]1$", colnames(x)) %>% which
x <- x[,keep.subj]
y <- colnames(x) %>%
strsplit("-") %>%
sapply(tail, 1) %>%
factor(levels=c("11", "01")) %>%
as.numeric
y.tab <- y %>% table %>% c
if ((length(y.tab) == 1) |
(!all(y.tab>10))) {
print("not enough subjects")
return()
}
N <- length(y)
BB <- 9999
## BB <- 9
seed.base <- 20200202 + ix.cancer*1e5
bootstrap.out <- list()
for (b in 1:BB) {
this.seed <- seed.base+b
set.seed(this.seed)
if ((b-1) %% 10 == 0) {
cat(sprintf("%s - % 6s % 5d\n", Sys.time(), cancer, b))
}
k <- 0
y.b <- rep(1, N)
y.b.tab <- y.b %>% table %>% c
while ((length(y.b.tab) == 1) | (!all(y.b.tab>10))) {
bootstrap.ix <- sample(N, replace=TRUE)
y.b <- y[bootstrap.ix]
y.b.tab <- y.b %>% table %>% c
k <- k+1
}
x.b <- x[,bootstrap.ix]
estimate_adj <- function(A, x, y) {
n_tags <- nrow(A)
node_tags <- rownames(A)
A1 <- A2 <- A
for (j in 1:n_tags) {
tag <- node_tags[j]
if (sum(A[j,]) > 0) {
parents_tags <- node_tags[as.logical(A[j,])]
correlation.data1 <- x[c(tag, parents_tags), y == 1, drop = FALSE] %>%
t %>% unname %>% data.frame
rm_ix1 <-
correlation.data1 %>%
apply(1, . %>% is.na %>% any) %>%
which
if (length(rm_ix1) > 0)
correlation.data1 <- correlation.data1[-rm_ix1,]
correlation.data2 <- x[c(tag, parents_tags), y == 2, drop = FALSE] %>%
t %>% unname %>% data.frame
rm_ix2 <-
correlation.data2 %>%
apply(1, . %>% is.na %>% any) %>%
which
if (length(rm_ix2) > 0)
correlation.data2 <- correlation.data2[-rm_ix2,]
cor.out1 <- pcor(correlation.data1)
cor.out2 <- pcor(correlation.data2)
partial.corr1 <- cor.out1[["estimate"]]
partial.corr2 <- cor.out2[["estimate"]]
rownames(partial.corr1) <- colnames(partial.corr1) <- c(tag, parents_tags)
rownames(partial.corr2) <- colnames(partial.corr2) <- c(tag, parents_tags)
A1[j, node_tags %in% parents_tags] <- partial.corr1[tag, parents_tags]
A2[j, node_tags %in% parents_tags] <- partial.corr2[tag, parents_tags]
}
}
A1[is.na(A1)] <- 0
A2[is.na(A2)] <- 0
return(list(A1 = A1, A2 = A2))
}
AA.miEMC <- estimate_adj(A.cancer, x.b, y.b)
subset.miEMC <- list(AA=AA.miEMC,
x=x.b)
make.subset <- function(regex, negate=TRUE, x, y) {
## Removes rows that match the regex
ix <- grepl(regex, rownames(x))
ix.row <- grepl(regex, rownames(A.cancer))
ix.col <- grepl(regex, colnames(A.cancer))
if (negate) {
ix <- not(ix)
ix.col <- not(ix.col)
ix.row <- not(ix.row)
}
xx <- x[ix,]
list(AA=estimate_adj(A.cancer[ix.row, ix.col],
xx, y),
x=xx)
}
subset.EMC <- make.subset("(^MIR|^LET|-mi$)", x=x.b, y=y.b)
subset.E <- make.subset("(^MIR|^LET|-mi$|-CN$|-ME$)", x=x.b, y=y.b)
subset.miE <- make.subset("(^MIR|^LET|-GE$)", negate=FALSE, x=x.b, y=y.b)
subsets <- list(miEMC=subset.miEMC,
miE=subset.miE,
EMC=subset.EMC,
E=subset.E)
all.gsa <- list()
for (i in 1:length(subsets)) {
## print(sprintf("%s", names(subsets)[i]))
AA <- subsets[[i]]$AA
xx <- subsets[[i]]$x
## A regular expression for any of the gene names, followed by "-GE"---eg,
## BRAF-GE
re.ge <- genes.A %>% paste(collapse="|") %>% sprintf("(%s)-GE", .)
## Any gene
re.genes <- genes.A %>% paste(collapse="|") %>% sprintf("(%s)-", .)
B <- matrix(c(1*grepl(re.ge, rownames(AA[[1]])),
1*grepl(re.genes, rownames(AA[[1]]))),
byrow=TRUE, nrow=2)
rownames(B) <- c("BRAF-GE", "BRAF-EMC")
gsa <- NetGSA(AA, xx, y.b, B=B, directed=TRUE, lklMethod="REML")
rownames(gsa$teststat) <- rownames(gsa$df) <- rownames(gsa$p.value) <- rownames(B)
all.gsa[[i]] <- gsa
}
names(all.gsa) <- names(subsets)
p <- do.call(cbind, lapply(all.gsa, function(g) g$p.value))
colnames(p) <- names(all.gsa)
T <- do.call(cbind, lapply(all.gsa, function(g) g$teststat))
colnames(T) <- names(all.gsa)
s2.e <- sapply(all.gsa, function(g) g$s2.epsilon)
s2.g <- sapply(all.gsa, function(g) g$s2.gamma)
beta1 <- list()
beta2 <- list()
for (gn in names(all.gsa)) {
gsa <- all.gsa[[gn]]
beta1[[gn]] <- gsa$beta[[1]]
beta2[[gn]] <- gsa$beta[[2]]
}
all.est <- list(beta1 = beta1,
beta2 = beta2,
s2.e = s2.e,
s2.g = s2.g,
p = p,
T = T)
gsa <- all.gsa
est <- all.est
bootstrap.out[[b]] <- list(bootstrap.ix=bootstrap.ix,
gsa=gsa,
est=est,
seed=this.seed)
}
## Save output from netgsa
save(bootstrap.out,
file=sprintf("out/%s/out-%s.Rdata", out.path, cancer))
}