forked from AmirhosseinModabbernia/ABCD
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathAnalysis_CCA
262 lines (220 loc) · 9.18 KB
/
Analysis_CCA
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
#########-----------------------------------------SCCA ANALYSIS- MORPHOMETRY---------------------------------------####
#THIS IS HOW ANALYSIS WAS DONE FOR THE MORPHOMETRIC DATA; SIMILAR ANALYSES WERE PERFORMED FOR RESTING STATE FUNCTIONAL CONNECTIVITY, GRAY-WHITE MATTER CONTRAST AND DIFFUSION WEIGHTED IMAGING DATASETS
#load packages
library(matrixStats)
library(mixOmics)
library(car)
library(parallel)
library(plyr)
library(dplyr)
library(MASS)
library(fabricatr)
library(vegan)
library(caret)
library(RANN)
library(CCA)
# reading data file
dftrain<-read.csv("dftrain_struct.csv")
dftrain<-as.matrix(dftrain[,-1])
dftest<-read.csv("dftest_struct.csv")
dftest<-as.matrix(dftest[,-1])
#loading permutation structures
permlist_struct_train<-read.csv("permutationlist_struct_train.csv",header = F)
permlist_struct_test<-read.csv("permutationlist_struct_test.csv",header = F)
Xtrain<-dftrain[,3:74]
Xtest<-dftest[,3:74]
Ytrain<-dftrain[,c(111:181,253:344)]
Ytest<-dftest[,c(111:181,253:344)]
ncomp=72
nperm=10000
#perform pca
pcaclin<-pca(Xtrain,ncomp)
pcaimag<-pca(Ytrain,159)
ncolx = ncol(pcaclin$variates$X)
ncoly = ncol(pcaimag$variates$X)
pcclintrainvar<-pcaclin$variates$X
pcimagtrainvar<-pcaimag$variates$X
pcclintrainweight<-pcaclin$loadings$X
pcimagtrainweight<-pcaimag$loadings$X
pcclintestvar<-Xtest%*%pcaclin$loadings$X
pcimagtestvar<-Ytest%*%pcaimag$loadings$X
#perform canonical correlation
x<-cc(pcclintrainvar,pcimagtrainvar)
#save all correlations
cor_struct_train<-diag(cor(x$scores$xscores,x$scores$yscores))
write.csv(cor_struct_train,"cor_struct_train.csv")
#calculate loadings
loadings_struct_train_clinical<-cor(Xtrain,x$scores$xscores,use="complete.obs")
write.csv(loadings_struct_train_clinical,"loadings_struct_train_clinical.csv")
loadings_struct_train_imaging<-cor(Ytrain,x$scores$yscores,use="complete.obs")
write.csv(loadings_struct_train_imaging,"loadings_struct_train_imaging.csv")
#save weights
weights_struct_train_clinical<-x$xcoef
weights_struct_train_imaging<-x$ycoef
#do permutations by using the feature weights from the original discovery set
numCores <- detectCores()
perm<-1:nperm
corperm_test<-matrix(0,nperm,ncomp)
corperm_train<-matrix(0,nperm,ncomp)
#test cor
cor_struct_test<-diag(cor(pcclintestvar%*%x$xcoef,pcimagtestvar%*%x$ycoef,use="complete.obs" ))
write.csv(cor_struct_test,"cor_struct_test.csv")
permfunc<-function(perm) {
testpsych = list()
testpsych$X1<-as.matrix(pcclintestvar[permlist_struct_test[,perm],])
testpsych$Y1<-as.matrix(pcimagtestvar)
corperm_test[perm,]<-diag(cor(testpsych$X1%*%x$xcoef,testpsych$Y1%*%x$ycoef,use="complete.obs" ))
}
results <- mclapply(perm, permfunc, mc.cores = numCores-10)
corperm_mat<-matrix(unlist(results),ncomp,nperm)
pval_struct_test<- c()
for (i in 1:ncomp) {
pval_struct_test[i] <-
length(which (corperm_mat[i, ] >=
cor_struct_test[i])) / nperm
}
write.csv(pval_struct_test,"pval_struct_test_BasedOnTrainWeights.csv")
#do permutations by using the feature weights on the permutated discovery set (on the test set)
numCores <- detectCores()
perm<-1:nperm
corperm_test<-matrix(0,nperm,ncomp)
corperm_train<-matrix(0,nperm,ncomp)
permfunc<-function(perm) {
temppsych = list()
temppsych$X1<-pcclintrainvar[permlist_struct_train[,perm],]
temppsych$Y1<-pcimagtrainvar
x<-cc(temppsych$X1,temppsych$Y1)
corperm_test[perm,]<-diag(cor(pcclintestvar%*%x$xcoef,pcimagtestvar%*%x$ycoef))
}
results <- mclapply(perm, permfunc, mc.cores = numCores-5)
corperm_mat<-matrix(unlist(results),ncomp,nperm)
pval_struct_test<- c()
for (i in 1:ncomp) {
pval_struct_test[i] <-
length(which (corperm_mat[i, ] >=
cor_struct_test[i])) / nperm
}
write.csv(pval_struct_test,"pval_struct_test_BasedOnPermutedTrainCCA.csv")
#do permutations by using the feature weights on the permutated discovery set (on the discovery set)
permfunc<-function(perm) {
temppsych = list()
temppsych$X1<-pcclintrainvar[permlist_struct_train[,perm],]
temppsych$Y1<-pcimagtrainvar
x<-cc(temppsych$X1,temppsych$Y1)
corperm_train[perm,]<-diag(cor(temppsych$X1%*%x$xcoef,temppsych$Y1%*%x$ycoef ))
}
results <- mclapply(perm, permfunc, mc.cores = numCores-5)
corperm_mat<-matrix(unlist(results),ncomp,nperm)
pval_struct_train<- c()
for (i in 1:ncomp) {
pval_struct_train[i] <-
length(which (corperm_mat[i, ] >=
cor_struct_train[i])) / nperm
}
write.csv(pval_struct_train,"pval_struct_train.csv")
#bootstrap resampling stability
set.seed(123)
perm=1:100
boot_func<-function(perm) {
cor_boot_struct<-matrix(0,15,1)
sampleseq<- c(.1,.2,.3,.4,.5,.6,.7,.8,.9,1,1.1,1.2,1.3,1.4,1.5)
for (j in sampleseq) {
trainsamp<-sample(nrow(dftrain),size=ceiling(j*nrow(dftrain)),replace=TRUE)
trainlist = list()
trainlist$X1 = pcclintrainvar[trainsamp,]
trainlist$Y1 = pcimagtrainvar[trainsamp,]
x<-cc(trainlist$X1,trainlist$Y1)
cor_boot_struct[which(sampleseq==j),]<-cor(x$scores$xscores,x$scores$yscores)[1,1]
}
return(cor_boot_struct)
}
results <- mclapply(perm, boot_func, mc.cores = numCores-5)
absresults<-lapply(results,abs)
struct_resample<-t(matrix(unlist(absresults),15,100))
write.csv(struct_resample,"struct_resample.csv")
#Leave one site out
origmat<-rbind(weights_struct_train_clinical,weights_struct_train_imaging)
#Leave one site out
loadingretest<-matrix(0,ncomp,22)
for ( site in 1:22) {
trainlist = list()
trainlist$X1 = pcclintrainvar[dftrain[,"site"]!=site, ]
trainlist$Y1 = pcimagtrainvar[dftrain[,"site"]!=site,]
x<-cc(trainlist$X1,trainlist$Y1)
#realign the new weight matrix based on the correlation with the original weight matrix
newmat<-rbind(x$xcoef,x$ycoef)
cormat=cor(origmat, newmat,method="spearman")
order<-c()
order[1]<-which(abs(cormat)==max(abs(cor(origmat,newmat[,1],method="spearman"))),arr.ind=TRUE)[[1]]
for ( i in 2:ncol(newmat)) {
k=1
while (which(abs(cormat)==sort(abs(cor(origmat,newmat[,i],method="spearman")),decreasing=TRUE)[k],arr.ind=T)[1] %in% order) {
k=k+1}
order[i]<-which(abs(cormat)==sort(abs(cor(origmat,newmat[,i],method="spearman")),decreasing=TRUE)[k],arr.ind=T)[1]
}
loadingretest[,site]<-abs(diag(cor(rbind(weights_struct_train_clinical,weights_struct_train_imaging),rbind(x$xcoef[,order],x$ycoef[,order]))))
}
write.csv(loadingretest,"leavesiteout_struct.csv")
numCores <- detectCores()
perm<-1:nperm
corperm_test<-matrix(0,nperm,ncomp)
corperm_train<-matrix(0,nperm,ncomp)
#test cor
cor_struct_test<-diag(cor(pcclintestvar%*%x$xcoef,pcimagtestvar%*%x$ycoef,use="complete.obs" ))
write.csv(cor_struct_test,"cor_struct_test.csv")
origmat<-rbind(weights_struct_train_clinical,weights_struct_train_imaging)
#split resampling reliability -test set
set.seed(123)
perm=1:1000
boot_func<-function(perm) {
#0.8 is being used instead of 0.85, because given than we are randomly sampling based on family IDs and not individuals, 0.8 gives us roughly 85% of the cases
trainsamp<-which(dftrain[,'family_id_final'] %in% sample(dftrain[,'family_id_final'],size=0.8*nrow(dftrain),replace=FALSE)==TRUE)
trainlist = list()
testlist=list()
trainlist$X1 = pcclintrainvar[trainsamp,]
trainlist$Y1 = pcimagtrainvar[trainsamp,]
x<-cc(trainlist$X1,trainlist$Y1)
newmat<-rbind(x$xcoef,x$ycoef)
testlist$X1 = pcclintrainvar[-trainsamp,]
testlist$Y1 = pcimagtrainvar[-trainsamp,]
cormat=cor(origmat, newmat,method="spearman")
order<-c()
order[1]<-which(abs(cormat)==max(abs(cor(origmat,newmat[,1],method="spearman"))),arr.ind=TRUE)[[1]]
for ( i in 2:ncol(newmat)) {
k=1
while (which(abs(cormat)==sort(abs(cor(origmat,newmat[,i],method="spearman")),decreasing=TRUE)[k],arr.ind=T)[1] %in% order) {
k=k+1}
order[i]<-which(abs(cormat)==sort(abs(cor(origmat,newmat[,i],method="spearman")),decreasing=TRUE)[k],arr.ind=T)[1]
}
corperm_train<-diag(cor(testlist$X1%*%x$xcoef[,order],testlist$Y1%*%x$ycoef[,order]))
}
results <- mclapply(perm, boot_func, mc.cores = numCores-5)
struct_resample<-t(matrix(unlist(results),ncomp,1000))
write.csv(struct_resample[,1],"cor8515test_structresample.csv")
#split resampling reliability discovery set
set.seed(123)
perm=1:1000
boot_func<-function(perm) {
trainsamp<-which(dftrain[,'family_id_final'] %in% sample(dftrain[,'family_id_final'],size=0.8*nrow(dftrain),replace=FALSE)==TRUE)
trainlist = list()
testlist=list()
trainlist$X1 = pcclintrainvar[trainsamp,]
trainlist$Y1 = pcimagtrainvar[trainsamp,]
x<-cc(trainlist$X1,trainlist$Y1)
newmat<-rbind(x$xcoef,x$ycoef)
testlist$X1 = pcclintrainvar[-trainsamp,]
testlist$Y1 = pcimagtrainvar[-trainsamp,]
cormat=cor(origmat, newmat,method="spearman")
order<-c()
order[1]<-which(abs(cormat)==max(abs(cor(origmat,newmat[,1],method="spearman"))),arr.ind=TRUE)[[1]]
for ( i in 2:ncol(newmat)) {
k=1
while (which(abs(cormat)==sort(abs(cor(origmat,newmat[,i],method="spearman")),decreasing=TRUE)[k],arr.ind=T)[1] %in% order) {
k=k+1}
order[i]<-which(abs(cormat)==sort(abs(cor(origmat,newmat[,i],method="spearman")),decreasing=TRUE)[k],arr.ind=T)[1]
}
corperm_train<-diag(cor(trainlist$X1%*%x$xcoef[,order],trainlist$Y1%*%x$ycoef[,order]))
}
results <- mclapply(perm, boot_func, mc.cores = numCores-5)
struct_resample<-t(matrix(unlist(results),ncomp,1000))
write.csv(struct_resample[,1],"cor8515train_structresample.csv")