-
Notifications
You must be signed in to change notification settings - Fork 0
/
8_zeitzeiger.R
269 lines (246 loc) · 12.8 KB
/
8_zeitzeiger.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
# FILE DESCRIPTION: 24_zeitzeiger ---------------------------------------------
#
# DESCRIPTION : Performs modelling of the physiological time based on gene
# expression with use of zeitzeiger
#
# USAGE:
#
# OPTIONS: none
# REQUIREMENTS:
# BUGS: --
# NOTES: ---
# AUTHOR: Maria Litovchenko, maria.litovchenko@epfl.ch
# COMPANY: EPFL, Lausanne, Switzerland
# VERSION: 1
# CREATED: 06.04.2018
# REVISION: 06.04.2018
setwd('~/Desktop/BitBucket/AroundTheClock_Aug_Oct2017/')
setwd('~/Desktop/AroundTheClock_Aug_Oct2017/')
source('0_common_functions.R')
# INPUTS ----------------------------------------------------------------------
# info about samples
INFO <- readRDS('Rds/sampleInfo.Rds')
INFO <- as.data.table(INFO)
setkey(INFO, rightGT_name)
# list all tissues and corresponding colors
allTissues <- unique(INFO$Tissue)
tissueColor <- c('darkorchid3', 'brown3', 'darkgoldenrod1', 'darkgreen')
names(tissueColor) <- allTissues
# output dirs
saveRDSdir <- 'Rds/'
savePlotsDir <- 'plots/'
saveBedDir <- 'beds/'
tissueOutliers <- list(FB = c('FB43_Oct_2', 'FB30_Oct_2', 'BRB2_FB_ZT11_29655'),
GUT = c('BRB5_Gut_ZT11_29655', 'BRB5_Gut_ZT23_29655',
'BRB5_Gut_ZT11_25205', 'BRB4_Gut_ZT9_28211',
'GUT31_Jan_6', 'BRB4_Gut_ZT21_29655'),
BRAIN = c('BRAIN1_Oct_3', 'BRAIN13_Oct_3',
'BRAIN28_Oct_1', 'BRAIN37_Oct_3',
'BRB7_Brain_ZT21_28211',
'BRB8_Brain_ZT1_29655'))
# WHITE & DGRP: READ-IN COUNTS, NORMALIZE, BATCH CORRECT ----------------------
# in this case, I put dgrps and w- into one table for normalization and
# batch correction, so there wouldn't be batch effect bacause of it
# prepare info table
tissue <- 'GUT'
whiteDGRP <- readRDS(paste0(saveRDSdir, 'whiteDGRP.Rds'))[[tissue]]
whiteDGRP <- whiteDGRP[, !colnames(whiteDGRP) %in% tissueOutliers[[tissue]]]
# separate into the training set + test set : white and profiled DGRPs
# and set of interest - DGRPs profiled once
whiteDgrpProf <- whiteDGRP[, !grepl('_n_', colnames(whiteDGRP))]
newRownames <- colnames(whiteDgrpProf)
whiteDgrpProf <- as.data.frame(t(whiteDgrpProf))
rownames(whiteDgrpProf) <- newRownames
dgrpNorm <- whiteDGRP[, grepl('_n_', colnames(whiteDGRP))]
newRownames <- colnames(dgrpNorm)
dgrpNorm <- as.data.frame(t(dgrpNorm))
rownames(dgrpNorm) <- newRownames
# time for both sets
whiteDgrpProfTime <- (INFO[rownames(whiteDgrpProf)]$Time %% 24 ) / 24
dgrpNormTime <- (INFO[rownames(dgrpNorm)]$Time %% 24 ) / 24
# STEP 1: best parameters -----------------------------------------------------
# To determine the best parameters for training a ZeitZeiger predictor, we can
# run cross-validation
registerDoParallel(cores = 4)
sumabsv <- c(1, 1.5, 3) # amount of regularization
nSpc <- 1:5 # how many SPCs are used for prediction
nFolds <- 10 # can only range from 1 to 10
foldid <- sample(rep(1:nFolds, length.out = nrow(whiteDgrpProf)))
# calculate cross-validation
# fit a periodic smoothing spline to the behavior of each feature as a function
# of time
fitCrossVal <- zeitzeigerFitCv(whiteDgrpProf, whiteDgrpProfTime, foldid)
spcCrossVal <- list()
predCrossVal <- list()
for (oneSumabsv in 1:length(sumabsv)) {
# fits to calculate sparse principal components (SPCs) for how the features
# change over time
spcCrossVal[[oneSumabsv]] <- zeitzeigerSpcCv(fitCrossVal,
sumabsv = sumabsv[oneSumabsv])
# uses the training data and the SPCs to predict the corresponding time for
# each test observation
predCrossVal[[oneSumabsv]] <- zeitzeigerPredictCv(as.matrix(whiteDgrpProf),
whiteDgrpProfTime,
foldid,
spcCrossVal[[oneSumabsv]],
nSpc = nSpc)
}
# reorganize the output, making a data.frame with the information for each
# prediction.
timePredCrossVal <- lapply(predCrossVal, function(x) x$timePred)
crossValRes <- data.frame(do.call(rbind, timePredCrossVal),
timeObs = rep(whiteDgrpProfTime, length(sumabsv)),
sumabsv = rep(sumabsv,
each = length(whiteDgrpProfTime)),
obs = rep(rownames(whiteDgrpProf), length(sumabsv)),
stringsAsFactors = F)
crossValResGath <- gather(crossValRes, key = nSpc, value = timePred, -obs,
-timeObs, -sumabsv)
crossValResGath$nSpc <- as.integer(sapply(as.character(crossValResGath$nSpc),
function(x) substr(x, 2, nchar(x))))
crossValResGath$sumabsv <- factor(crossValResGath$sumabsv)
crossValResGath$timeError <- calcTimeDiff(crossValResGath$timeObs,
crossValResGath$timePred)
# median absolute error for each set of parameter values.
crossValResGathGroup <- crossValResGath %>% group_by(sumabsv, nSpc) %>%
summarize(medae = median(abs(timeError)))
# Plot the error for each set of parameter values
ggplot(crossValResGathGroup) +
geom_point(aes(x = nSpc, y = medae, shape = sumabsv, color = sumabsv),
size=2) +
labs(x = 'Number of SPCs', y = 'Median absolute error') +
theme_bw() + theme(legend.position = c(0.7, 0.7))
# !!! the best accuracy seems to be at sumabsv = 1.5 and nSpc = 2
optSumabsv <- 1.5
optSPC <- 2
timeDiffCutOff <- 5
# STEP 2: Train a model on white- and profiled DGRPs --------------------------
# fit a periodic smoothing spline to the behavior of each feature as a function
# of time
fitWhiteDgrpProf <- zeitzeigerFit(whiteDgrpProf, whiteDgrpProfTime)
# fits to calculate sparse principal components (SPCs) for how the features
# change over time
spcWhiteDgrpProf <- zeitzeigerSpc(fitWhiteDgrpProf$xFitMean,
fitWhiteDgrpProf$xFitResid,
sumabsv = optSumabsv)
# uses the training data and the SPCs to predict the corresponding time for
# each test observation
predfitWhiteDgrpProf <- zeitzeigerPredict(as.matrix(whiteDgrpProf),
whiteDgrpProfTime,
as.matrix(whiteDgrpProf),
spcWhiteDgrpProf, nSpc = optSPC)
# compare predicted and harversted time
harvPredTime <- data.frame(timeObs = whiteDgrpProfTime,
timePred = predfitWhiteDgrpProf$timePred,
timeError = calcTimeDiff(whiteDgrpProfTime,
predfitWhiteDgrpProf$timePred))
plot(24 * harvPredTime[, 1:2], pch = 20, bty = 'n',
col = ifelse(abs(harvPredTime$timeError) > timeDiffCutOff / 24,
tissueColor[tissue], 'black'), cex = 2,
xlim = c(0, 25), ylim = c(0, 25), cex.lab = 2, cex.axis = 2,
xlab = 'Time of harversting, h', ylab = 'Physiological time, h',
main = paste0('Physiological and harversting time, ', tissue))
abline(1, 1, lty = 2)
text(24 * harvPredTime[abs(harvPredTime$timeError) > timeDiffCutOff / 24, 1:2],
labels = rownames(whiteDgrpProf)[abs(harvPredTime$timeError) >
timeDiffCutOff / 24],
pos = 3)
grid()
ggplot(harvPredTime, aes(x = 24 * timeObs, y = 24 * timeError)) +
geom_point(size = 2, shape = 1) +
scale_x_continuous(limits = 24 * c(0, 1)) +
labs(x = 'Observed time', y = 'Error') + theme_bw()
# plot proporion of explained variance by each of SPC
dfVar <- data.frame(spc = 1:length(spcWhiteDgrpProf$d),
propVar = spcWhiteDgrpProf$d^2 / sum(spcWhiteDgrpProf$d^2))
ggplot(dfVar) + geom_point(aes(x = spc, y = propVar), size = 2, shape = 1) +
scale_x_continuous(breaks = seq(1, 10)) +
labs(x = 'SPC', y = 'Proportion of\nvariance explained') + theme_bw()
# genes which are in SPCs
genesPredictors <- colnames(whiteDgrpProf)[which(rowSums(spcWhiteDgrpProf$v[, 1:optSPC]) != 0)]
message(paste("Genes-predictors:", paste(genesPredictors, collapse = ',')))
spcNot0Matr <- spcWhiteDgrpProf$v[rowSums(spcWhiteDgrpProf$v[, 1:optSPC]) != 0, ][, 1:optSPC]
sapply(1:optSPC,
function(x) message(paste('SPC', x, ':',
paste(genesPredictors[which(spcNot0Matr[, x] != 0)],
collapse = ', '))))
# See, which genes drive the predictions: project the observations from
# feature-space to SPC-space, to look at how the optimal SPCs behave over time
spcMatr <- as.matrix(whiteDgrpProf) %*% spcWhiteDgrpProf$v[, 1:optSPC]
colnames(spcMatr) <- paste('SPC', 1:optSPC)
zGath <- gather(data.frame(spcMatr, obs = rownames(whiteDgrpProf),
Time = whiteDgrpProfTime, check.names = F),
key = SPC, value = Abundance, -obs, -Time)
ggplot(zGath) + facet_grid(SPC ~ ., scales = 'free_y') +
geom_point(aes(x = Time, y = Abundance), size = 2, shape = 1) +
ylab('Gene(s) expression') + theme_bw()
# Plot the coefficients of the genes for the SPCs
spcNot0Matr <- as.data.frame(spcNot0Matr)
colnames(spcNot0Matr) <- paste('SPC', 1:optSPC)
spcNot0Matr$Gene <- genesPredictors
spcNot0MatrGath <- gather(spcNot0Matr, key = spc,
value = Coefficient, -Gene) %>%
mutate(feature = factor(Gene, levels = rev(spcNot0Matr$Gene)))
ggplot(spcNot0MatrGath) + facet_wrap(~ spc, nrow = 1) +
geom_bar(aes(x = Gene, y = Coefficient), stat = 'identity') +
labs(x = 'Gene') + coord_flip() +
theme_bw() + theme(panel.spacing = unit(1.2, 'lines'))
# STEP 3: Predict time for DGRPs profiled once --------------------------------
optSumabsv <- 1.5
optSPC <- 2
timeDiffCutOff <- 5
whiteDGRP <- readRDS(paste0(saveRDSdir, 'whiteDGRP.Rds'))
for (tissue in allTissues) {
whiteDGRPtiss <- whiteDGRP[[tissue]]
whiteDGRPtiss <- whiteDGRPtiss[, !colnames(whiteDGRPtiss) %in% "" ]# tissueOutliers[[tissue]]]
# separate into the training set + test set : white and profiled DGRPs
# and set of interest - DGRPs profiled once
whiteDgrpProftiss <- whiteDGRPtiss[, !grepl('_n_', colnames(whiteDGRPtiss))]
newRownames <- colnames(whiteDgrpProftiss)
whiteDgrpProftiss <- as.data.frame(t(whiteDgrpProftiss))
rownames(whiteDgrpProftiss) <- newRownames
dgrpNormtiss <- whiteDGRPtiss[, grepl('_n_', colnames(whiteDGRPtiss))]
newRownames <- colnames(dgrpNormtiss)
dgrpNormtiss <- as.data.frame(t(dgrpNormtiss))
rownames(dgrpNormtiss) <- newRownames
# time for both sets
whiteDgrpProfTissTime <- (INFO[rownames(whiteDgrpProftiss)]$Time %% 24 ) / 24
dgrpNormTissTime <- (INFO[rownames(dgrpNormtiss)]$Time %% 24 ) / 24
# fit a periodic smoothing spline to the behavior of each feature as a function
# of time
fitWhiteDgrpProfTiss <- zeitzeigerFit(whiteDgrpProftiss,
whiteDgrpProfTissTime)
# fits to calculate sparse principal components (SPCs) for how the features
# change over time
spcWhiteDgrpProfTiss <- zeitzeigerSpc(fitWhiteDgrpProfTiss$xFitMean,
fitWhiteDgrpProfTiss$xFitResid,
sumabsv = optSumabsv)
# uses the training data and the SPCs to predict the corresponding time for
# each test observation
predfitDGRPsTiss <- zeitzeigerPredict(as.matrix(whiteDgrpProftiss),
whiteDgrpProfTissTime,
as.matrix(dgrpNormtiss),
spcWhiteDgrpProfTiss, nSpc = optSPC)
# compare predicted and harversted time
harvPredTimeDGRPsTiss <- data.frame(Sample = rownames(dgrpNormtiss),
timeObs = 24 * dgrpNormTissTime,
timePred = 24 * predfitDGRPsTiss$timePred)
harvPredTimeDGRPsTiss$timeError <- apply(harvPredTimeDGRPsTiss, 1,
function(x) diffInTime(as.numeric(x[2]),
as.numeric(x[3])))
plot(harvPredTimeDGRPsTiss[, 2:3], pch = 20, bty = 'n',
col = ifelse(abs(harvPredTimeDGRPsTiss$timeError) > timeDiffCutOff,
tissueColor[tissue], 'black'), cex = 2,
xlim = c(0, 25), ylim = c(0, 25), cex.lab = 2, cex.axis = 2,
xlab = 'Time of harversting, h', ylab = 'Physiological time, h',
main = paste0('Physiological and harversting time, ', tissue))
abline(1, 1, lty = 2)
text(harvPredTimeDGRPsTiss[abs(harvPredTimeDGRPsTiss$timeError) >
timeDiffCutOff, 2:3],
labels = rownames(dgrpNormtiss)[abs(harvPredTimeDGRPsTiss$timeError) >
timeDiffCutOff],
pos = 3)
grid()
saveRDS(harvPredTimeDGRPsTiss, paste0(saveRDSdir, tissue,
'_physioPredTimeDF_ZZ.Rds'))
}