-
Notifications
You must be signed in to change notification settings - Fork 3
/
Copy pathlasso_subtype_meta.R
360 lines (287 loc) · 13.7 KB
/
lasso_subtype_meta.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
# Author: Jeffrey Thompson
# Purpose: Performs multiclass lasso logistic regression to classify cancer subtypes.
# Get command line arguments.
#args <- commandArgs(trailingOnly = T)
source("package_loader.R")
source("directories.R")
NFOLDS = 100
NSEEDS = 10
INITIAL_SEED = 2
ACCURACY_PLOT_FILENAME = "meta_accuracies.pdf"
KAPPA_PLOT_FILENAME = "meta_kappas.pdf"
ACCURACY_TABLE_FILENAME = "meta_accuracies.tsv"
BALANCED_PLOT_FILENAME = "meta_balanced_accuracies.pdf"
args = c(normalized_data,
'META_BRCA_ZEROONE.pcl',
'combined_subtype.txt',
'META_BRCA_MA_ZEROONE.pcl',
'BRCAClin.tsv',
'META_BRCA_LOG_ZEROONE.pcl',
'BRCARNASeqClin.tsv',
'META_BRCA_QN_ZEROONE.pcl',
'BRCARNASeqClin.tsv',
'META_BRCA_TDM_ZEROONE.pcl',
'BRCARNASeqClin.tsv',
'META_BRCA_NPN_ZEROONE.pcl',
'BRCARNASeqClin.tsv',
'META_BRCA_REF_NPN_ZEROONE.pcl',
'combined_subtype.txt',
'META_BRCA_UN_ZEROONE.pcl',
'BRCARNASeqClin.tsv',
output)
input_dir = args[1]
ref_input = args[2]
ref_clin = args[3]
tcga_input = args[4]
tcga_clin = args[5]
log_input = args[6]
log_clin = args[7]
qn_input = args[8]
qn_clin = args[9]
tdm_input = args[10]
tdm_clin = args[11]
npn_input = args[12]
npn_clin = args[13]
ref_npn_input = args[14]
ref_npn_clin = args[15]
un_input = args[16]
un_clin = args[17]
output_dir = args[18]
load_it(c("glmnet", "caret", "e1071", "stringr", "plyr", "huge"))
# This function creates dataframes to hold the accuracies for each subtype across runs.
setup_acc_df = function(df) {
df = data.frame(matrix(nrow=5, ncol=0))
rownames(df) = c("Basal", "Her2", "LumA", "LumB", "Normal")
return (df)
}
# Define data.frames to hold classification results.
metadf = setup_acc_df(metadf)
tcgadf = setup_acc_df(tcgadf)
logdf = setup_acc_df(logdf)
qndf = setup_acc_df(qndf)
tdmdf = setup_acc_df(tdmdf)
npndf = setup_acc_df(npndf)
undf = setup_acc_df(undf)
tdmnotshareddf = setup_acc_df(tdmnotshareddf)
qnnotshareddf = setup_acc_df(qnnotshareddf)
lognotshareddf = setup_acc_df(lognotshareddf)
npnnotshareddf = setup_acc_df(npnnotshareddf)
unnotshareddf = setup_acc_df(unnotshareddf)
# Define random seeds.
set.seed(INITIAL_SEED)
seeds = sample(1:10000, NSEEDS)
message(paste("Random seeds:", paste(seeds, collapse=", ")), appendLF=TRUE)
# Pre-processes test data to ready it for the LASSO.
# Returns a list with first entry the pre-processed dataset and second entry
# the classes for those data.
preprocessTCGA = function(dataFile, clinFile, RNAseq=TRUE) {
data = read.table(paste0(input_dir, dataFile),
sep='\t',
row.names=1,
header=T)
dataT = t(data)
# Remove duplicate samples from the data.
dataT = dataT[str_detect(substr(rownames(dataT),1,15), "(\\.01)|(\\.11)"),]
# Read clinical data.
clin = read.table(paste0(input_dir, clinFile), sep='\t', row.names=1, header=T)
if(!RNAseq) {
clin[,1] = clin[,3]
}
# Remove samples without subtype labels.
clin = subset(clin, clin[,1] != "")
clin[,1] = droplevels(clin[,1])
# Filter clinical data to include only those samples with expression data.
clin.response = clin[,1][match(substr(chartr('-', '.', rownames(dataT)), 1, 15), substr(chartr('-', '.', rownames(clin)),1,15))]
clin.response = na.omit(clin.response)
# Filter expression data to include only those samples with ids also in clinical data.
dataT = dataT[substr(chartr('-', '.', rownames(dataT)), 1, 15) %in% substr(chartr('-', '.', rownames(clin)),1,15),]
# Filter any data with missing clinical annotation for tumor subtype.
dataT = dataT[!is.na(clin[,1][match(substr(chartr('-', '.', rownames(dataT)),1,15),
substr(chartr('-', '.', rownames(clin)),1,15))]),]
return (list(data=dataT, response=clin.response))
} # end function preprocessTCGA
# Predict subtypes on TCGA data using previously trained model.
predictTCGA = function(title, data, model) {
# Predict classes based on the trained model.
lasso.predict = predict(model, data$data, s = "lambda.1se", type="class")
# Make sure all factors are included.
lasso.predict = factor(lasso.predict, c("Basal", "Her2", "LumA", "LumB", "Normal"))
# Build a contingency table of the results.
con_table = table(pred = lasso.predict, true = data$response, exclude="none")
# Generate statistics based on the contigency table.
cm = confusionMatrix(con_table)
return (cm)
}
# Load the training data.
train = read.table(paste0(input_dir, ref_input), sep='\t', row.names=1, header=T)
# Transpose rows and columns.
trainT = t(train)
# Load the clinical features.
clin = read.table(paste0(input_dir, ref_clin), sep='\t', row.names=1, header=T)
# Remove annotations that have no subtype label.
clin = subset(clin, clin[,1] != "NC")
clin[,1] = droplevels(clin[,1])
# Filter the expression data to remove those samples without subtype labels.
trainT = trainT[rownames(trainT) %in% chartr('-', '.', rownames(clin)),]
# Filter clinical data to include only those entries that also have expression data.
trainclin.subtypes = clin[,1][match(chartr('-', '.', rownames(trainT)), chartr('-', '.', rownames(clin)))]
trainclin.subtypes = na.omit(trainclin.subtypes)
write.table(trainclin.subtypes, file=paste0(output_dir, "meta_ref_classes.txt"))
# Filter any expression data with missing clinical annotation for tumor subtype.
trainT = trainT[!is.na(clin[,1][match(chartr('-', '.', rownames(trainT)), chartr('-', '.', rownames(clin)))]),]
# Load the training data.
npn_train = read.table(paste0(input_dir, ref_npn_input), sep='\t', row.names=1, header=T)
# Transpose rows and columns.
npntrainT = t(npn_train)
# Filter the expression data to remove those samples without subtype labels.
npntrainT = npntrainT[rownames(npntrainT) %in% chartr('-', '.', rownames(clin)),]
# Filter any expression data with missing clinical annotation for tumor subtype.
npntrainT = npntrainT[!is.na(clin[,1][match(chartr('-', '.', rownames(npntrainT)), chartr('-', '.', rownames(clin)))]),]
# Pre-process the TCGA data.
tcga = preprocessTCGA(tcga_input, tcga_clin, FALSE)
log = preprocessTCGA(log_input, log_clin)
qn = preprocessTCGA(qn_input, qn_clin)
tdm = preprocessTCGA(tdm_input, tdm_clin)
npn = preprocessTCGA(npn_input, npn_clin)
un = preprocessTCGA(un_input, un_clin)
# Determine which samples are shared between TCGA RNA-seq and TCGA microarray data.
# Then filter the data to include only those that are shared.
tdm_shared = substr(chartr('-', '.', rownames(tdm$data)),1,15) %in% substr(chartr('-', '.', rownames(tcga$data)),1,15)
tdm$data = tdm$data[tdm_shared,]
tdm$response = tdm$response[tdm_shared]
tcga_shared = substr(chartr('-', '.', rownames(tcga$data)),1,15) %in% substr(chartr('-', '.', rownames(tdm$data)),1,15)
tcga$data = tcga$data[tcga_shared,]
tcga$response = tcga$response[tcga_shared]
qn_shared = substr(chartr('-', '.', rownames(qn$data)),1,15) %in% substr(chartr('-', '.', rownames(tcga$data)),1,15)
qn$data = qn$data[qn_shared,]
qn$response = qn$response[qn_shared]
log_shared = substr(chartr('-', '.', rownames(log$data)),1,15) %in% substr(chartr('-', '.', rownames(tcga$data)),1,15)
log$data = log$data[log_shared,]
log$response = log$response[log_shared]
npn_shared = substr(chartr('-','.', rownames(npn$data)),1,15) %in% substr(chartr('-','.',rownames(tcga$data)),1,15)
npn$data = npn$data[npn_shared,]
npn$response = npn$response[npn_shared]
un_shared = substr(chartr('-', '.', rownames(un$data)),1,15) %in% substr(chartr('-','.',rownames(tcga$data)),1,15)
un$data = un$data[un_shared,]
un$response = un$response[un_shared]
# Pre-allocate vectors to hold results.
tcga_accs = vector(mode="list", length=length(seeds))
tdm_accs = vector(mode="list", length=length(seeds))
qn_accs = vector(mode="list", length=length(seeds))
log_accs = vector(mode="list", length=length(seeds))
npn_accs = vector(mode="list", length=length(seeds))
un_accs = vector(mode="list", length=length(seeds))
# Perform a number of iterations equal to the number of random seeds.
# At each iteration, perform n-fold cross validation to build a model on the training data,
# then use that model to make predictions on the test data.
for(seednum in 1:length(seeds)) {
set.seed(seeds[seednum])
# Train a model for classifying cancer subtype on the METABRIC training data.
set.seed(seeds[seednum])
lasso.model=cv.glmnet(trainT, trainclin.subtypes, family="multinomial", parallel = F, type.measure="class", nfolds=NFOLDS)
plot(lasso.model)
set.seed(seeds[seednum])
npn.lasso.model=cv.glmnet(npntrainT, trainclin.subtypes, family="multinomial", parallel=F, type.measure="class", nfolds=NFOLDS)
acc = predictTCGA("TCGA Results", tcga, lasso.model)
tcga_accs[[seednum]] = acc
tcgadf = cbind(tcgadf, as.vector(acc$byClass[,8]))
acc <- predictTCGA("TDM Results", tdm, lasso.model)
tdm_accs[[seednum]] = acc
tdmdf <- cbind(tdmdf,as.vector(acc$byClass[,8]))
acc <- predictTCGA("QN Results", qn, lasso.model)
qn_accs[[seednum]] = acc
qndf <- cbind(qndf,as.vector(acc$byClass[,8]))
acc <- predictTCGA("LOG Results", log, lasso.model)
log_accs[[seednum]] = acc
logdf <- cbind(logdf,as.vector(acc$byClass[,8]))
acc = predictTCGA("NPN Results", npn, npn.lasso.model)
npn_accs[[seednum]] = acc
npndf = cbind(npndf,as.vector(acc$byClass[,8]))
acc = predictTCGA("Untransformed Results", un, lasso.model)
un_accs[[seednum]] = acc
undf = cbind(undf,as.vector(acc$byClass[,8]))
}
# Build a table of accuracies across all datasets.
accuracies = data.frame(Basal=numeric(0), Her2=numeric(0), LumA=numeric(0), LumB=numeric(0), Normal=numeric(0))
accuracies = rbind(accuracies, apply(tcgadf, 1, mean))
accuracies = rbind(accuracies, apply(tdmdf,1,mean))
accuracies = rbind(accuracies, apply(qndf,1,mean))
accuracies = rbind(accuracies, apply(logdf,1,mean))
accuracies = rbind(accuracies, apply(npndf,1,mean))
accuracies = rbind(accuracies, apply(undf,1,mean))
rownames(accuracies) <- c("MA", "TDM", "QN", "LOG", "NPN", "UNTR")
colnames(accuracies) <- c("Basal", "Her2", "LumA", "LumB", "Normal")
# Aggregate accuracies:
tdm_tables = lapply(tdm_accs, function(x) x$table)
qn_tables = lapply(qn_accs, function(x) x$table)
log_tables = lapply(log_accs, function(x) x$table)
tcga_tables = lapply(tcga_accs, function(x) x$table)
npn_tables = lapply(npn_accs, function(x) x$table)
un_tables = lapply(un_accs, function(x) x$table)
tdm_reduced = Reduce("+", tdm_tables) / length(tdm_tables)
qn_reduced = Reduce("+", qn_tables) / length(qn_tables)
log_reduced = Reduce("+", log_tables) / length(log_tables)
tcga_reduced = Reduce("+", tcga_tables) / length(tcga_tables)
npn_reduced = Reduce("+", npn_tables) / length(npn_tables)
un_reduced = Reduce("+", un_tables) / length(un_tables)
tdm_reduced[1:5,1:5] = t(apply(tdm_reduced, 1, function(x) as.integer(round(x))))[1:5,1:5]
qn_reduced[1:5,1:5] = t(apply(qn_reduced, 1, function(x) as.integer(round(x))))[1:5,1:5]
log_reduced[1:5,1:5] = t(apply(log_reduced, 1, function(x) as.integer(round(x))))[1:5,1:5]
tcga_reduced[1:5,1:5] = t(apply(tcga_reduced, 1, function(x) as.integer(round(x))))[1:5,1:5]
npn_reduced[1:5,1:5] = t(apply(npn_reduced, 1, function(x) as.integer(round(x))))[1:5,1:5]
un_reduced[1:5,1:5] = t(apply(un_reduced, 1, function(x) as.integer(round(x))))[1:5,1:5]
tdm_cm = confusionMatrix(tdm_reduced)
qn_cm = confusionMatrix(qn_reduced)
log_cm = confusionMatrix(log_reduced)
tcga_cm = confusionMatrix(tcga_reduced)
npn_cm = confusionMatrix(npn_reduced)
un_cm = confusionMatrix(un_reduced)
saveRDS(log_cm, file=paste0(output_dir, "meta_log_cm.RDS"))
all_accs = data.frame(rbind(tdm_cm$overall, qn_cm$overall, log_cm$overall, npn_cm$overall, un_cm$overall, tcga_cm$overall))
all_accs = cbind(all_accs, method=c("TDM", "QN", "LOG", "NPN", "UNTR", "MA"))
nir = all_accs[1,]$AccuracyNull
all_accs$method = factor(all_accs$method, levels=all_accs$method)
# Plot accuracies
ci = aes(ymin=AccuracyLower, ymax=AccuracyUpper)
cbPalette = c("#000000", "#E69F00", "#56B4E9", "#009E73", "#F0E442", "#0072B2", "#D55E00", "#CC79A7")
ggplot(all_accs, aes(x=factor(method), y=Accuracy, color=method)) +
geom_point(size=3) +
geom_errorbar(ci, width=.3) +
ylab("Total Accuracy") +
xlab("Normalization") +
theme_bw() +
theme(axis.text.x = element_text(angle = 90, hjust = 1)) +
geom_hline(aes(yintercept = nir), linetype="longdash") +
scale_color_manual(values=cbPalette)
ggsave(paste0(output_dir, ACCURACY_PLOT_FILENAME), plot=last_plot(), width=3, height=2.5)
cbPalette = c("#000000", "#E69F00", "#56B4E9", "#009E73", "#F0E442", "#0072B2", "#D55E00", "#CC79A7")
ggplot(data = all_accs) +
ylab("Kappa") +
xlab("Normalization") +
theme_bw() +
theme(axis.text.x = element_text(angle = 90, hjust = 1)) +
geom_point(aes(x=factor(method), y=Kappa, color=method), shape=18, size=4) +
scale_color_manual(values=cbPalette)
ggsave(paste0(output_dir, KAPPA_PLOT_FILENAME), plot=last_plot(), width=3, height=2.5)
meta_all = rbind(tdm_cm$byClass[,8],
qn_cm$byClass[,8],
log_cm$byClass[,8],
npn_cm$byClass[,8],
un_cm$byClass[,8],
tcga_cm$byClass[,8])
meta_melted = melt(meta_all)
colnames(meta_melted) = c("Dataset", "Subtype", "Accuracy")
meta_melted$Dataset = c("TDM", "QN", "LOG", "NPN", "UNTR", "MA")
meta_melted$Dataset = factor(meta_melted$Dataset, levels=c("TDM", "QN", "LOG", "NPN", "UNTR", "MA"))
cbPalette = c("#000000", "#E69F00", "#56B4E9", "#009E73", "#F0E442", "#0072B2", "#D55E00", "#CC79A7")
ggplot(data=meta_melted, aes(x=Dataset, y=Accuracy, ymax=1)) +
ylim(0.45,1) +
coord_flip() +
geom_point(size=3, aes(colour=Dataset)) +
theme(axis.text.x = element_text(angle = 90, hjust = 1)) +
theme(plot.title=element_text(hjust=0)) +
ylab("Balanced Accuracy") + theme_bw() + facet_wrap(~Subtype, ncol=5) +
theme(legend.position="none") +
scale_color_manual(values=cbPalette)
ggsave(paste0(output_dir, BALANCED_PLOT_FILENAME), plot=last_plot(), height=2.5, width=6)
write.table(all_accs, file=paste0(output_dir, ACCURACY_TABLE_FILENAME), sep='\t', row.names=T, col.names=T)