-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy path2way_EBADIMEXgenes_predict.R
463 lines (404 loc) · 27.6 KB
/
2way_EBADIMEXgenes_predict.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
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
#!/usr/bin/env Rscript
library(optparse)
# Command Line Arguments
option_list <- list(
make_option(c("-d","--data"), type='character', help = 'cached data to use', default = "data/RData/data_LUSC_all_lazyReady.R_CACHE"),
make_option(c("-r","--results"), type='character', help = 'model results to use', default = "data/latestResults/top_100_each_seed.RData"),
make_option(c("-f","--folds"), type='integer', help = 'number of folds to split the dataset', default = 4),
make_option(c("-n","--name"), type='character', help = 'name of the analysis performed', default = "MLs_EBADIMEX_50randomSamples_BRCA_progressing"),
make_option("--n_top", type='integer', help = 'number of top ranks/genes to train for and combine', default = 100),
make_option(c("-p","--parallel"), type='integer', help = 'if desired, provide a number of cores to parallelize the execution of the analysis', default = 1)
)
#opt <- parse_args(OptionParser(option_list = option_list))
opt <- list(data = "data/RData/data_BRCA_progressing_lazyReady.R_CACHE", results="data/latestResults/top_100_each_seed.RData", name = "PINCAGE_EBADIMEX_50randomSamples_BRCA_progressing", folds = 50, n_top = 100, parallel=8)
library(caret)
library(pROC)
library(dplyr)
library(parallel)
source("R/utilities_new.R")
load(opt$results)
load("/home/michal/iCancerGenomics/irksome-fibula/essentials_EBADIMEXgenes.RData")
load("/home/michal/iCancerGenomics/irksome-fibula/UQs_BRCA_progressing.RData")
samplenames <- unlist(read_grouping(cache_name = opt$data, group1="progressed", group2="nonProgressed"))
indiv_predictions_PINCAGE_cv <- matrix(ncol=length(samplenames), nrow=opt$n_top)
colnames(indiv_predictions_PINCAGE_cv) <- samplenames
# configure here
res_pr <- 25 # number of bins for promoter methylation
res_gb <- 25 # number of bins for gene body methylation
res_expr <- 25 # number of bins for gene expression
smooth_1d <- 1 # should expression node parameterization be smoothed?
smooth_2d <- 1 # should methylation 2D factors parameterizations be smoothed?
if (smooth_1d > 0) library(aws)
if (smooth_2d > 0) library(smoothie)
# the following objects are expected by this script: workingList_BRCA, mmatrix_pc, counts_plusOne, G1, G2, factors_ls, TSS1500Ind, TSS200Ind, UTR5Ind, EXON1Ind, GENEBODYInd, UTR3Ind
args <- commandArgs(trailingOnly = TRUE)
beg <- as.numeric(args[1]) # first ID to process
end <- as.numeric(args[2]) # last consequitve ID to process
integrand_e <- function(x, k) {dpois(k, x)}
integrand_m <- function(x, mean) {dnorm(x=mean, mean=x, sd=0.14)}
tensor_product <- function(matrix1, matrix2, smooth_h=0, normalize=c("row", "column", "no"),kernel=c("gauss", "cauchy", "minvar")) {
if (is.matrix(matrix1) && is.matrix(matrix2)) result <- matrix(ncol=ncol(matrix1),nrow=ncol(matrix2),data=rep(0,ncol(matrix1)*ncol(matrix2))) else result <- matrix(ncol=length(matrix1),nrow=length(matrix1),data=rep(0,length(matrix1)*length(matrix2)))
if (is.matrix(matrix1) && is.matrix(matrix2)) for (i in 1:nrow(matrix1)) {
result <- result + matrix(ncol=ncol(matrix1),nrow=ncol(matrix2),byrow=TRUE,data=apply(expand.grid(matrix1[i,],matrix2[i,]), 1, prod))
} else result <- result + matrix(nrow=length(matrix1),ncol=length(matrix2),byrow=TRUE,data=apply(expand.grid(matrix1,matrix2), 1, prod))
if (is.matrix(matrix1) && is.matrix(matrix2)) result <- result/nrow(matrix1)
if (!is.null(kernel) && smooth_h > 0) result <- kernel2dsmooth(result,kernel.type = kernel[1],sigma=smooth_h,nx=ncol(matrix2),ny=ncol(matrix1))
if (normalize[1] == "row") for (i in 1:nrow(result)) result[i,] <- result[i,]/sum(result[i,]) else if (normalize[1] == "column") for (i in 1:nrow(result)) result[,i] <- result[,i]/sum(result[,i])
return(result)
}
geo_mean <- function(data) {
log_data <- log(data)
gm <- exp(mean(log_data[is.finite(log_data)]))
return(gm)
}
sum_asLogs <- function (x) {
if (x[1] >= x[2]) x[1] + log(1+ exp(x[2]-x[1])) else x[2] + log(1+ exp(x[1]-x[2]))
}
AUCs <- NULL
CV_folds <- NULL
for (cv_run in 1:50) {
current_genes <- top_100_each_seed[[cv_run]]
system(command=paste('mkdir ', cv_run, sep=""))
# for (gene in 1:length(current_genes)) {
if (opt$parallel > 1) {
out <- mclapply(1:length(current_genes), mc.cores = opt$parallel, FUN = function(gene) {
cat(paste("doing run ", cv_run,", ", gene, "\n",sep=""))
#ptm <- proc.time()[3]
set.seed(cv_run)
train_set_g1 <- sample(1:38, 10)
train_set_g2 <- 38+sample(1:43, 15)
test_set_g1 <- (1:38)[-train_set_g1]
test_set_g2 <- 38+(1:43)[-(train_set_g2-38)]
genedata <- as.data.frame(read_genedata(cache = opt$data, genename = current_genes[gene])) %>%
dplyr::select(starts_with("pr_"), starts_with("gb_"), matches("read_count"))
genedata_training <- genedata %>%
slice(c(train_set_g1, train_set_g2))
genedata_testing <- genedata %>%
slice(c(test_set_g1, test_set_g2))
G1 <- rownames(genedata)[1:38]
G2 <- rownames(genedata)[39:81]
train_set_g1 <- rownames(genedata)[train_set_g1]
train_set_g2 <- rownames(genedata)[train_set_g2]
test_set_g1 <- rownames(genedata)[test_set_g1]
test_set_g2 <- rownames(genedata)[test_set_g2]
rownames(genedata_training) <- c(train_set_g1, train_set_g2)
rownames(genedata_testing) <- c(test_set_g1, test_set_g2)
system(command=paste('mkdir ', cv_run, '/', gene, sep=""))
system(command=paste('mkdir ./', cv_run, '/', gene, '/G1_model',sep=""))
system(command=paste('mkdir ./', cv_run, '/', gene, '/G2_model',sep=""))
###########################################################
############### identify constitutive CpGs ################
IDs_promoter <- colnames(genedata)[grep("pr_", colnames(genedata))]
promoterVars <- promoterVars_template[1:length(IDs_promoter)]
promoter_CpGs <- template_promoter_CpGs[1:length(IDs_promoter)]
IDs_body <- colnames(genedata)[grep("gb_", colnames(genedata))]
geneBodyVars <- geneBodyVars_template[1:length(IDs_body)]
geneBody_CpGs <- template_body_CpGs[1:length(IDs_body)]
ncol = length(IDs_promoter) + length(IDs_body) + 1
############################################################
######### calculate epsilons and smoothing params ##########
epsilon_pr_G2 <- 1/(length(train_set_g2)*length(IDs_promoter))/res_pr
epsilon_gb_G2 <- 1/(length(train_set_g2)*length(IDs_body))/res_gb
epsilon_e_G2 <- 1/length(train_set_g2)/res_expr
epsilon_pr_G1 <- 1/(length(train_set_g1)*length(IDs_promoter))/res_pr
epsilon_gb_G1 <- 1/(length(train_set_g1)*length(IDs_body))/res_gb
epsilon_e_G1 <- 1/length(train_set_g1)/res_expr
smooth_e <- 10/(mean(length(train_set_g1),length(train_set_g2))/(res_expr*2))
smooth_pr <- trunc(10/(mean(length(train_set_g1),length(train_set_g2))*length(IDs_promoter)/(res_pr*res_expr)))
smooth_gb <- trunc(10/(mean(length(train_set_g1),length(train_set_g2))*length(IDs_body)/(res_gb*res_expr)))
###########################################################
############### define the binning scheme ################
all_labels_pr <- as.character(seq(1,res_pr,1))
all_labels_gb <- as.character(seq(1,res_gb,1))
all_labels_expr <- as.character(seq(1,res_expr,1))
pseudo_counts_pr <- matrix(ncol=res_pr,nrow=res_expr,data=rep(1,res_expr*res_pr))
pseudo_counts_gb <- matrix(ncol=res_gb,nrow=res_expr,data=rep(1,res_expr*res_gb))
# gene body
x <- t(genedata[c(G2,G1), IDs_body])
density <- density(x,bw=0.14,from=-7,to=7,n=2801,na.rm=TRUE)
density$y <- density$y/sum(density$y)
density$y <- cumsum(density$y)
breaks <- NULL
noBreaks <- res_gb-1
for (j in 1:noBreaks) { breaks <- c (breaks, density$x[which(density$y >= j*(1/(1+noBreaks)))][1])}
breaksBODY <- sort(c(-7.01,breaks,7.01))
# promoter
x <- t(genedata[c(G2,G1), IDs_promoter])
density <- density(x,bw=0.14,from=-7,to=7,n=2801,na.rm=TRUE)
density$y <- density$y/sum(density$y)
density$y <- cumsum(density$y)
breaks <- NULL
noBreaks <- res_pr-1
for (j in 1:noBreaks) { breaks <- c (breaks, density$x[which(density$y >= j*(1/(1+noBreaks)))][1])}
breaksPROMOTER <- sort(c(-7.01,breaks,7.01))
# expression
temp_UQs <- matrix(ncol=2)
colnames(temp_UQs) <- c("UQ","density")
for (j in 1:nrow(genedata)) {
lambda <- as.numeric(genedata$read_count[j]) + 1
X <- seq(round(max(lambda - (4 * lambda * lambda^(-1/2)), 1)), round(lambda + (4 * lambda * lambda^(-1/2))))
current <- UQs[c(G2, G1)[j]]
temp_UQs <- rbind(temp_UQs, cbind(X / current, dpois(X, lambda = lambda) * current))
}
temp_UQs <- as.data.frame(temp_UQs[-1,],)
temp_UQs <- temp_UQs[order(temp_UQs$UQ),]
temp_UQs[,3] <- cumsum(temp_UQs[,2])
temp_UQs[,3] <- temp_UQs[,3]/max(temp_UQs[,3])
breaks <- NULL
noBreaks <- res_expr-1
for (j in 1:noBreaks) { breaks <- c (breaks, temp_UQs[which(temp_UQs[,3] >= j*(1/(1+noBreaks))),1][1])}
breaksEXPRESSION <- c(0,breaks,10^6)
########################################################
# dynamic generation of model specification files here #
########################################################
####################### state map ######################
stateMaps <- file(paste("./",cv_run, '/', gene, "/stateMaps.txt", sep=""),"w")
exprMap <- paste("NAME:\texprMap\nSYMBOLS:\t", paste(seq(1, res_expr, 1), collapse=" "), "\nMETA_SYMBOLS:\t.=", paste(seq(1, res_expr, 1), collapse=" "), "; *=", paste(seq(1, res_expr, 1), collapse=" "), ";\n\n", collapse="", sep="")
prMap <- paste("NAME:\tprMap\nSYMBOLS:\t", paste(seq(1, res_pr, 1), collapse=" "), "\nMETA_SYMBOLS:\t.=" ,paste(seq(1, res_pr, 1), collapse=" "), "; *=", paste(seq(1, res_pr, 1), collapse=" "), ";\n\n", collapse="", sep="")
gbMap <- paste("NAME:\tgbMap\nSYMBOLS:\t", paste(seq(1, res_gb, 1),collapse=" "), "\nMETA_SYMBOLS:\t.=", paste(seq(1, res_gb, 1), collapse=" "), "; *=", paste(seq(1, res_gb, 1), collapse=" "), ";\n", collapse="", sep="")
cat(exprMap, prMap, gbMap, file=stateMaps)
close(stateMaps)
########################################################
####################### variables ######################
variables <- file(paste("./", cv_run, '/', gene, "/variables.txt",sep=""),"w")
cat("STATE_MAP_NAME:\texprMap\nVAR_NAMES:\tEXPR\n\n",sep="",file=variables)
cat("STATE_MAP_NAME:\tprMap\nVAR_NAMES:\tM.P\n\n",sep="",file=variables)
cat("STATE_MAP_NAME:\tgbMap\nVAR_NAMES:\tM.GB\n",sep="",file=variables)
close(variables)
########################################################
##################### factor graph #####################
factorGraph <- file(paste("./", cv_run, '/', gene,"/factorGraph.txt", sep=""), "w")
cat("NAME:\tEXPR.likelihood\nNB1:\tEXPR\nPOT:\tpot_EXPR.likelihood\n", file=factorGraph)
cat("\nNAME:\tEXPR.prior\nNB1:\tEXPR\nPOT:\tpot_EXPR.prior\n", file=factorGraph)
cat(paste("\nNAME:\tEXPR.M.GB\nNB1:\tEXPR\nNB2:\tM.GB\nPOT:\tpot_EXPR.M.GB\n", sep="", collapse=""), file=factorGraph)
cat(paste("\nNAME:\tEXPR.M.P\nNB1:\tEXPR\nNB2:\tM.P\nPOT:\tpot_EXPR.M.P\n", sep="", collapse=""), file=factorGraph)
cat(paste("\nNAME:\t", promoter_CpGs, "\nNB1:\tM.P\nPOT:\tpot_", promoter_CpGs,"\n",sep="", collapse=""), file=factorGraph)
cat(paste("\nNAME:\t", geneBody_CpGs, "\nNB1:\tM.GB\nPOT:\tpot_", geneBody_CpGs,"\n",sep="", collapse=""), file=factorGraph)
close(factorGraph)
system(command=paste('cp ./',cv_run, '/', gene,'/*.txt ./',cv_run, '/', gene,'/G1_model/',sep=""))
system(command=paste('cp ./',cv_run, '/', gene,'/*.txt ./',cv_run, '/', gene,'/G2_model/',sep=""))
# initialize parameter precomputation matrices
promoter_G2 <- matrix(ncol=res_pr,nrow=length(G2))
promoter_G1 <- matrix(ncol=res_pr,nrow=length(G1))
body_G2 <- matrix(ncol=res_gb,nrow=length(G2))
body_G1 <- matrix(ncol=res_gb,nrow=length(G1))
expr_G2 <- matrix(ncol=res_expr,nrow=length(G2))
expr_G1 <- matrix(ncol=res_expr,nrow=length(G1))
rownames(promoter_G2) <- rownames(body_G2) <- rownames(expr_G2) <- G2
rownames(promoter_G1) <- rownames(body_G1) <- rownames(expr_G1) <- G1
###########################################################################
################### G1 model using training data ##########################
###########################################################################
# generate FacData
tempS_G1 <- matrix(ncol=ncol, nrow=length(G1))
rownames(tempS_G1) <- G1
for (current_sample in 1:length(G1)) {
# expression
read_count <- as.numeric(trunc(genedata[G1[current_sample], "read_count"])) # adding a pseudo-count to avoide problems with 0-count data
lambdas <- breaksEXPRESSION * UQs[G1[current_sample]]
frequencies_expr <- rep(0,length(breaksEXPRESSION)-1)
for (freq in 1:res_expr) {
frequencies_expr[freq] <- integrate(integrand_e, lower = lambdas[freq], upper = lambdas[freq+1], read_count,stop.on.error=FALSE)[1]
}
frequencies_expr <- unlist(frequencies_expr)
if (length(which(frequencies_expr==0))==res_expr) frequencies_expr[length(frequencies_expr)] <- 1
frequencies_expr <- frequencies_expr + epsilon_e_G1
frequencies_expr <- frequencies_expr/sum(frequencies_expr)
# gene body
cpg_list_gb <- NULL
for (cpg in 1:length(IDs_body)) {
miu <- genedata[G1[current_sample], IDs_body[cpg]]
if (!is.na(miu)) {
frequencies_gb <- rep(0,res_gb)
for (freq in 1:res_gb) {
frequencies_gb[freq] <- integrate(integrand_m,lower=breaksBODY[freq],upper=breaksBODY[freq+1],mean=miu)$value
}
frequencies_gb <- unlist(frequencies_gb) + epsilon_gb_G1
frequencies_gb <- frequencies_gb/sum(frequencies_gb)
cpg_list_gb[[cpg]] <- frequencies_gb
} else cpg_list_gb[[cpg]] <- rep(1/res_gb,res_gb)
}
# promoter
cpg_list_pr <- NULL
for (cpg in 1:length(IDs_promoter)) {
miu <- genedata[G1[current_sample], IDs_promoter[cpg]]
if (!is.na(miu)) {
frequencies_pr <- rep(0,res_pr)
for (freq in 1:res_pr) {
frequencies_pr[freq] <- integrate(integrand_m,lower=breaksPROMOTER[freq],upper=breaksPROMOTER[freq+1],mean=miu)$value
}
frequencies_pr <- unlist(frequencies_pr) + epsilon_pr_G1
frequencies_pr <- frequencies_pr/sum(frequencies_pr)
cpg_list_pr[[cpg]] <- frequencies_pr
} else cpg_list_pr[[cpg]] <- rep(1/res_pr,res_pr)
}
tempS_formated <- matrix(ncol=ncol, nrow=1)
tempS_formated[1,1] <- paste('[1,', res_expr,']((', paste(frequencies_expr,sep="", collapse=","), '))', sep="", collapse="")
cur_ncol <- 1
for (element in 1:length(cpg_list_pr)) {
tempS_formated[1, cur_ncol+1] <- paste('[1,', res_pr,']((', paste(cpg_list_pr[[element]], sep="", collapse=","),'))', sep="", collapse="")
cur_ncol <- cur_ncol + 1
}
for (element in 1:length(cpg_list_gb)) {
tempS_formated[1, cur_ncol+1] <- paste('[1,', res_gb,']((', paste(cpg_list_gb[[element]], sep="", collapse=","), '))', sep="", collapse="")
cur_ncol <- cur_ncol + 1
}
tempS_G1[current_sample,] <- tempS_formated
#start precomputing correct initialization of parameters
promoter_G1[current_sample,] <- apply(matrix(unlist(cpg_list_pr), ncol=res_pr, byrow=TRUE), 2,geo_mean)/sum(apply(matrix(unlist(cpg_list_pr), ncol=res_pr,byrow=TRUE),2, geo_mean))
body_G1[current_sample,] <- apply(matrix(unlist(cpg_list_gb), ncol=res_gb, byrow=TRUE), 2, geo_mean)/sum(apply(matrix(unlist(cpg_list_gb),ncol=res_gb,byrow=TRUE), 2, geo_mean))
expr_G1[current_sample,] <- frequencies_expr
}
##########################################################################
################### G2 model using training data #########################
##########################################################################
# generate FacData
tempS_G2 <- matrix(ncol=ncol, nrow=length(G2))
rownames(tempS_G2) <- G2
for (current_sample in 1:length(G2)) {
# expression
read_count <- as.numeric(trunc(genedata[G2[current_sample], "read_count"])) + 1 # adding a pseudo-count to avoide problems with 0-count data
lambdas <- breaksEXPRESSION * UQs[G2[current_sample]]
frequencies_expr <- rep(0,length(breaksEXPRESSION)-1)
for (freq in 1:res_expr) {
frequencies_expr[freq] <- integrate(integrand_e, lower = lambdas[freq], upper = lambdas[freq+1], read_count,stop.on.error=FALSE)[1]
}
frequencies_expr <- unlist(frequencies_expr)
if (length(which(frequencies_expr==0))==res_expr) frequencies_expr[length(frequencies_expr)] <- 1
frequencies_expr <- frequencies_expr + epsilon_e_G2
frequencies_expr <- frequencies_expr/sum(frequencies_expr)
# gene body
cpg_list_gb <- NULL
for (cpg in 1:length(IDs_body)) {
miu <- genedata[G2[current_sample], IDs_body[cpg]]
if (!is.na(miu)) {
frequencies_gb <- rep(0,res_gb)
for (freq in 1:res_gb) {
frequencies_gb[freq] <- integrate(integrand_m,lower=breaksBODY[freq],upper=breaksBODY[freq+1],mean=miu)$value
}
frequencies_gb <- unlist(frequencies_gb) + epsilon_gb_G2
frequencies_gb <- frequencies_gb/sum(frequencies_gb)
cpg_list_gb[[cpg]] <- frequencies_gb
} else cpg_list_gb[[cpg]] <- rep(1/res_gb,res_gb)
}
# promoter
cpg_list_pr <- NULL
for (cpg in 1:length(IDs_promoter)) {
miu <- genedata[G2[current_sample], IDs_promoter[cpg]]
if (!is.na(miu)) {
frequencies_pr <- rep(0,res_pr)
for (freq in 1:res_pr) {
frequencies_pr[freq] <- integrate(integrand_m,lower=breaksPROMOTER[freq],upper=breaksPROMOTER[freq+1],mean=miu)$value
}
frequencies_pr <- unlist(frequencies_pr) + epsilon_pr_G2
frequencies_pr <- frequencies_pr/sum(frequencies_pr)
cpg_list_pr[[cpg]] <- frequencies_pr
} else cpg_list_pr[[cpg]] <- rep(1/res_pr,res_pr)
}
tempS_formated <- matrix(ncol=ncol, nrow=1)
tempS_formated[1,1] <- paste('[1,',res_expr,']((',paste(frequencies_expr,sep="",collapse=","),'))',sep="",collapse="")
cur_ncol <- 1
for (element in 1:length(cpg_list_pr)) {
tempS_formated[1,cur_ncol+1] <- paste('[1,',res_pr,']((',paste(cpg_list_pr[[element]],sep="",collapse=","),'))',sep="",collapse="")
cur_ncol <- cur_ncol + 1
}
for (element in 1:length(cpg_list_gb)) {
tempS_formated[1,cur_ncol+1] <- paste('[1,',res_gb,']((',paste(cpg_list_gb[[element]],sep="",collapse=","),'))',sep="",collapse="")
cur_ncol <- cur_ncol + 1
}
tempS_G2[current_sample,] <- tempS_formated
#start precomputing correct initialization of parameters
promoter_G2[current_sample,] <- apply(matrix(unlist(cpg_list_pr), ncol=res_pr, byrow=TRUE), 2,geo_mean)/sum(apply(matrix(unlist(cpg_list_pr), ncol=res_pr, byrow=TRUE), 2, geo_mean))
body_G2[current_sample,] <- apply(matrix(unlist(cpg_list_gb), ncol=res_gb, byrow=TRUE), 2,geo_mean)/sum(apply(matrix(unlist(cpg_list_gb), ncol=res_gb, byrow=TRUE), 2, geo_mean))
expr_G2[current_sample,] <- frequencies_expr
}
########################################
# build the G1 model using training data
########################################
prior_pr <- apply(promoter_G1[train_set_g1,], 2, mean)
prior_gb <- apply(body_G1[train_set_g1,], 2, mean)
if (smooth_1d > 0) {
prior_expr <- kernsm(apply(expr_G1[train_set_g1,], 2, mean), h=smooth_e)
prior_expr <- prior_expr@yhat/sum(prior_expr@yhat)
} else prior_expr <- apply(expr_G1, 2, mean)
string <- paste(prior_pr, collapse=",")
promoterPots <- paste("\nNAME:\t\tpot_", promoter_CpGs, "\nTYPE:\t\trowNorm\nPOT_MAT:\t[1,", res_pr,"]((", string, "))\nPC_MAT:\t\t[1,", res_pr, "]((",paste(rep(1, res_pr), collapse=","), "))\n", sep="", collapse="")
string <- paste(prior_gb,collapse=",")
geneBodyPots <- paste("\nNAME:\t\tpot_", geneBody_CpGs, "\nTYPE:\t\trowNorm\nPOT_MAT:\t[1,", res_gb, "]((", string, "))\nPC_MAT:\t\t[1,", res_gb,"]((",paste(rep(1, res_gb), collapse=","), "))\n", sep="" , collapse="")
string <- paste(prior_expr, collapse=",")
expr.pots <- paste("\nNAME:\t\tpot_EXPR.likelihood\nTYPE:\t\trowNorm\nPOT_MAT:\t[1,", res_expr, "]((", string, "))\nPC_MAT:\t\t[1,", res_expr, "]((", paste(rep(1,res_expr), collapse=","), "))\n\nNAME:\t\tpot_EXPR.prior\nTYPE:\t\trowNorm\nPOT_MAT:\t[1,", res_expr, "]((", string, "))\nPC_MAT:\t\t[1,", res_expr,"]((", paste(rep(1, res_expr), collapse=","),"))\n\n", sep="", collapse="")
result <- tensor_product(body_G1[train_set_g1,], expr_G1[train_set_g1,], smooth_h=smooth_gb)
expr.m <- paste("NAME:\t\tpot_EXPR.M.GB\nTYPE:\t\trowNorm\nPOT_MAT:\t\t[", paste(c(res_expr,res_gb), collapse=","), "]((" ,paste(apply(result , 1, paste, collapse=","), collapse="),\n\t\t\t(") ,"))\nPC_MAT:\t\t[" ,paste(c(res_expr, res_gb), collapse=","),"]((", paste(apply(pseudo_counts_gb, 1, paste, collapse=","), collapse="),\n\t\t\t("), "))\n\n", sep="", collapse="")
result <- tensor_product(promoter_G1[train_set_g1,], expr_G1[train_set_g1,], smooth_h=smooth_pr)
expr.m <- c(expr.m, paste("NAME:\t\tpot_EXPR.M.P\nTYPE:\t\trowNorm\nPOT_MAT:\t\t[", paste(c(res_expr, res_pr), collapse=","), "]((", paste(apply(result, 1, paste, collapse=","), collapse="),\n\t\t\t("), "))\nPC_MAT:\t\t[", paste(c(res_expr,res_pr), collapse=","), "]((", paste(apply(pseudo_counts_pr,1,paste, collapse=","), collapse="),\n\t\t\t("), "))\n\n", sep="", collapse=""))
potentials <- file(paste("./", cv_run, '/', gene, "/G1_model/factorPotentials.txt", sep=""), "w")
cat(expr.m, expr.pots, promoterPots, geneBodyPots, file=potentials)
close(potentials)
########################################
# build the G2 model using training data
########################################
prior_pr <- apply(promoter_G2[train_set_g2,], 2, mean)
prior_gb <- apply(body_G2[train_set_g2,], 2, mean)
if (smooth_1d > 0) {
prior_expr <- kernsm(apply(expr_G2[train_set_g2,], 2, mean), h=smooth_e)
prior_expr <- prior_expr@yhat/sum(prior_expr@yhat)
} else prior_expr <- apply(expr_G2[train_set_g2,], 2, mean)
string <- paste(prior_pr, collapse=",")
promoterPots <- paste("\nNAME:\t\tpot_", promoter_CpGs, "\nTYPE:\t\trowNorm\nPOT_MAT:\t[1,", res_pr,"]((", string,"))\nPC_MAT:\t\t[1,", res_pr,"]((", paste(rep(1,res_pr), collapse=","),"))\n", sep="", collapse="")
string <- paste(prior_gb, collapse=",")
geneBodyPots <- paste("\nNAME:\t\tpot_", geneBody_CpGs, "\nTYPE:\t\trowNorm\nPOT_MAT:\t[1,", res_gb,"]((", string,"))\nPC_MAT:\t\t[1,", res_gb,"]((", paste(rep(1, res_gb),collapse=",") ,"))\n", sep="", collapse="")
string <- paste(prior_expr, collapse=",")
expr.pots <- paste("\nNAME:\t\tpot_EXPR.likelihood\nTYPE:\t\trowNorm\nPOT_MAT:\t[1,", res_expr, "]((", string,"))\nPC_MAT:\t\t[1,", res_expr, "]((", paste(rep(1,res_expr), collapse=","),"))\n\nNAME:\t\tpot_EXPR.prior\nTYPE:\t\trowNorm\nPOT_MAT:\t[1,",res_expr, "]((", string, "))\nPC_MAT:\t\t[1,", res_expr, "]((", paste(rep(1, res_expr), collapse=","), "))\n\n", sep="", collapse="")
result <- tensor_product(body_G2, expr_G2, smooth_h=smooth_gb)
expr.m <- paste("NAME:\t\tpot_EXPR.M.GB\nTYPE:\t\trowNorm\nPOT_MAT:\t\t[", paste(c(res_expr, res_gb), collapse=","), "]((", paste(apply(result, 1, paste, collapse=","), collapse="),\n\t\t\t("),"))\nPC_MAT:\t\t[", paste(c(res_expr, res_gb), collapse=","), "]((", paste(apply(pseudo_counts_gb, 1, paste, collapse=","), collapse="),\n\t\t\t("), "))\n\n", sep="", collapse="")
result <- tensor_product(promoter_G2, expr_G2, smooth_h=smooth_pr)
expr.m <- c(expr.m, paste("NAME:\t\tpot_EXPR.M.P\nTYPE:\t\trowNorm\nPOT_MAT:\t\t[", paste(c(res_expr, res_pr), collapse=","),"]((", paste(apply(result, 1, paste, collapse=","), collapse="),\n\t\t\t("),"))\nPC_MAT:\t\t[", paste(c(res_expr,res_pr), collapse=","),"]((", paste(apply(pseudo_counts_pr, 1, paste, collapse=","), collapse="),\n\t\t\t("), "))\n\n", sep="", collapse=""))
potentials <- file(paste("./", cv_run, '/', gene, "/G2_model/factorPotentials.txt", sep=""), "w")
cat(expr.m, expr.pots, promoterPots, geneBodyPots, file=potentials)
close(potentials)
#################################################################
###### generate "missing" Var data for test samples #############
#################################################################
tempVar_G1 <- matrix(rep(".", length(test_set_g1) * 3), nrow=length(test_set_g1), ncol=3)
tempVar_G2 <- matrix(rep(".", length(test_set_g2)*3), nrow=length(test_set_g2), ncol=3)
colnames(tempVar_G2) <- colnames(tempVar_G1) <- c("NAME:\tEXPR", "M.GB", "M.P")
rownames(tempVar_G1) <- test_set_g1
rownames(tempVar_G2) <- test_set_g2
eval(parse(text = paste('write.table(', paste('tempVar_G1, file = "./', cv_run, '/', gene, '/G1_model/G1_VarData_test.tab", row.names=TRUE, col.names=TRUE, quote=FALSE, sep="\t", append=FALSE)', sep = ""))))
eval(parse(text = paste('write.table(', paste('tempVar_G2,file = "./', cv_run, '/', gene, '/G2_model/G2_VarData_test.tab", row.names=TRUE, col.names=TRUE, quote=FALSE,sep="\t", append=FALSE)', sep = ""))))
#################################################################
######## Write the G1 and G2 FacData for test samples ###########
#################################################################
tempFac <- tempS_G1[test_set_g1,]
colnames(tempFac) <- c("NAME:\tEXPR.likelihood", promoter_CpGs, geneBody_CpGs)
eval(parse(text = paste('write.table(', paste('tempFac, file ="./', cv_run, '/', gene,'/G1_model/G1_FacData_test.tab", row.names=TRUE, col.names=TRUE, quote=FALSE,sep="\t", append=FALSE)', sep = ""))))
tempFac <- tempS_G2[test_set_g2,]
colnames(tempFac) <- c("NAME:\tEXPR.likelihood", promoter_CpGs, geneBody_CpGs)
eval(parse(text = paste('write.table(', paste('tempFac,file ="./', cv_run, '/', gene,'/G2_model/G2_FacData_test.tab", row.names=TRUE, col.names=TRUE, quote=FALSE, sep="\t", append=FALSE)', sep = ""))))
###### Query the G1 and G2 with test samples ###########
# query the G1 model with G1 test samples
string<-system(intern=TRUE,command=paste('./dfgEval_static --dfgSpecPrefix=./', cv_run, '/', gene, '/G1_model/ -l -n - ./', cv_run, '/', gene, '/G1_model/G1_VarData_test.tab ./', cv_run, '/', gene, '/G1_model/G1_FacData_test.tab', sep=""))
G1test_G1model_mlogliks <- as.numeric(substring(string[-1], 17))
# query the G2 model with G2 test samples
string<-system(intern=TRUE,command=paste('./dfgEval_static --dfgSpecPrefix=./', cv_run, '/', gene, '/G2_model/ -l -n - ./', cv_run, '/', gene,'/G2_model/G2_VarData_test.tab ./', cv_run, '/', gene, '/G2_model/G2_FacData_test.tab', sep=""))
G2test_G2model_mlogliks <- as.numeric(substring(string[-1], 17))
# query the G1 model with G2 test samples
string<-system(intern=TRUE, command=paste('./dfgEval_static --dfgSpecPrefix=./', cv_run, '/', gene, '/G1_model/ -l -n - ./', cv_run, '/', gene,'/G2_model/G2_VarData_test.tab ./', cv_run, '/', gene,'/G2_model/G2_FacData_test.tab', sep=""))
G2test_G1model_mlogliks <- as.numeric(substring(string[-1], 17))
# query the G2 model with G1 test samples
string<-system(intern=TRUE, command=paste('./dfgEval_static --dfgSpecPrefix=./', cv_run, '/', gene,'/G2_model/ -l -n - ./', cv_run, '/', gene,'/G1_model/G1_VarData_test.tab ./', cv_run, '/', gene,'/G1_model/G1_FacData_test.tab', sep=""))
G1test_G2model_mlogliks <- as.numeric(substring(string[-1], 17))
samples_G1_likelihoods <- c(G1test_G1model_mlogliks, G2test_G1model_mlogliks)
samples_G2_likelihoods <- c(G1test_G2model_mlogliks, G2test_G2model_mlogliks)
# G1 posterior probability calculation
out <- data.frame(sample_ID=c(test_set_g1, test_set_g2), posterior_G1=exp(-samples_G1_likelihoods) / (exp(-samples_G1_likelihoods) + exp(-samples_G2_likelihoods)), G1_mloglik=samples_G1_likelihoods, G2_mloglik=samples_G2_likelihoods)
colnames(out) <- c("sample_ID", "posterior_G1", "G1_mloglik", "G2_mloglik")
out
})
# transfer the results from parallel execution
currentGenes_out <- NULL
for (gene in 1:length(out)) currentGenes_out[[gene]] <- out[[gene]]
CV_folds[[cv_run]] <- currentGenes_out
} else stop("Serial execution currently unavailable, use --parallel >1")
}
save(CV_folds, file=paste(opt$name, ".RData", sep=""))