/
ascca_CV_gamma_command_line.R
executable file
·214 lines (162 loc) · 8.71 KB
/
ascca_CV_gamma_command_line.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
#options(echo = FALSE)
library(MASS)
library(stats)
source("func.R")
# Import x.data
# Import y.data
args <- commandArgs(trailingOnly = TRUE)
x.data = read.table(args[1])
x.map = x.data[,1]
y.data = read.table(args[2])
y.map = y.data[,1]
#rows are observations, columns are genes(variables)
x.data = t(x.data[,-1])
y.data = t(y.data[,-1])
n.cv <- 5 # Number of cross-validation steps to select sparseness parameters
p <- length(x.map) # Number of varibles in X
q <- length(y.map) # Number of varibles in Y
n.sample <- nrow(x.data) # Sample size (# of observations or chips)
### _______________________________________________________________________________________
### Setting sparseness parameters
### _______________________________________________________________________________________
max.v = 0.4
max.u = 0.4
step.lambda = 0.01
min.gamma = 0 # start of gamma
max.gamma = 2 # end of gamma
step.gamma = 0.1
lambda.v.seq <- seq(0, max.v, by=step.lambda) # Possible values of sparseness parameters for data Y. Lower bouns should be 0, upper bound can be increased to 0.2.
lambda.u.seq <- seq(0, max.u, by=step.lambda) # Possible values of sparseness parameters for data X. Lower bouns should be 0, upper bound can be increased to 0.2.
gamma.seq <- seq(min.gamma, max.gamma, by=step.gamma) # Possible values of gammas.
n.lambdas.u <- length(lambda.u.seq)
n.lambdas.v <- length(lambda.v.seq)
n.gamma <- length(gamma.seq)
#lambda.v.matrix <- matrix(rep(lambda.v.seq, n.lambdas.u), nrow=n.lambdas.u, byrow=T)
lambda.v.matrix <- matrix(rep(lambda.v.seq, n.lambdas.u), nrow=n.lambdas.u, byrow=T)
lambda.u.matrix <- matrix(rep(lambda.u.seq, n.lambdas.v), nrow=n.lambdas.u, byrow=F)
lambda.v.array = array(rep(lambda.v.matrix,n.gamma),c(n.lambdas.u,n.lambdas.v,n.gamma))
lambda.u.array = array(rep(lambda.u.matrix,n.gamma),c(n.lambdas.u,n.lambdas.v,n.gamma))
gamma.array <- array(as.numeric(gl(n.gamma,n.lambdas.u*n.lambdas.v)),c(n.lambdas.u,n.lambdas.v,n.gamma))
ones.p <- rep(1, p)/p
ones.q <- rep(1, q)/q
### _______________________________________________________________________________________
### Analysis
### _______________________________________________________________________________________
out.file = "out_ascca_CV_gamma.txt"
#out.file = paste("ascca_CV_gamma_",n.cv,"_",max.u,"_",max.v,"_",step.lambda,"_",min.gamma,"_",max.gamma,"_",step.gamma,".txt",sep = "")
cat(date(),"\n",file = out.file)
cat("n.cv, max.u, max.v, step.lambda, min.gamma, max.gamma and step.gamma are:\n",n.cv,max.u,max.v,step.lambda,min.gamma,max.gamma,step.gamma,"\n",file = out.file,append = TRUE)
cat("begin select sparseness parameters:\n",file = out.file,append = TRUE)
#cat("begin select sparseness parameters:\n",file = "ascca_CV_gamma_iteration.txt",append = TRUE)
n.cv.sample <- trunc(n.sample/n.cv)
whole.sample <- seq(1, n.sample)
predict.corr.scca <- array(0, c(n.lambdas.u, n.lambdas.v, n.gamma)) # This array will contain average test sample correlation for each combination of sparseness parameters and gamma
#_______Cross-validation to select optimal combination of sparseness parameters____________
for (i.cv in 1:n.cv)
{
cat("cross validation",i.cv,":\n",file = out.file,append = TRUE)
testing.sample <- whole.sample[((i.cv-1)*n.cv.sample+1):(i.cv*n.cv.sample)]
training.sample <- whole.sample[!whole.sample%in%testing.sample]
k <- sample.sigma12.function(x.data[training.sample, ], y.data[training.sample, ])
# Get starting values for singular vectors
# as column and row means from matrix K
u.initial <- k %*% ones.q
u.initial <- u.initial /sqrt(as.numeric(t(u.initial)%*%u.initial))
v.initial <- t(k) %*% ones.p
v.initial <- v.initial /sqrt(as.numeric(t(v.initial)%*%v.initial))
# _______________Data for Predicted correlation (testing sample)_________________
x.predict <- x.data[testing.sample, ]
y.predict <- y.data[testing.sample, ]
# Standardize data
x.predict <- x.predict - mean(x.predict)
y.predict <- y.predict - mean(y.predict)
sigma11.predict <- var(x.predict)
sigma22.predict <- var(y.predict)
x.predict <- x.predict %*% diag( 1/sqrt(diag(sigma11.predict)) )
y.predict <- y.predict %*% diag( 1/sqrt(diag(sigma22.predict)) )
uv.svd = svd(k,nu=1,nv=1)
u.svd = uv.svd$u
v.svd = uv.svd$v
for(j.gamma in 1:n.gamma)
{
gamma = gamma.seq[j.gamma]
cat("when gamma = ",gamma,"\n",file = out.file,append = TRUE)
# ____________Loops for sparseness parameter combinations__________
for(j.lambda.v in 1:n.lambdas.v)
{
flag.na <- 0
for(j.lambda.u in 1:n.lambdas.u)
{
lambda.v <- lambda.v.seq[j.lambda.v] # sparseness parameter for Y
lambda.u <- lambda.u.seq[j.lambda.u] # sparseness parameter for X
if(flag.na==0)
{
uv <- adaptive.scca.function(k, u.initial, v.initial, lambda.u, lambda.v, u.svd, v.svd, gamma)
vj <- uv$v.new
uj <- uv$u.new
# Calculate predicted correlation for SCCA
predict.corr.scca[j.lambda.u, j.lambda.v, j.gamma] <- predict.corr.scca[j.lambda.u, j.lambda.v, j.gamma] + abs(cor(x.predict%*%uj, y.predict%*%vj))
#when either uj or vj or both are zero vector
if(is.na(predict.corr.scca[j.lambda.u, j.lambda.v, j.gamma]))
{
flag.na <- 1
cat("NA at",lambda.u,"and",lambda.v,"\n",file = out.file,append = TRUE)
#cat(uj,"\n",file = out.file,append = TRUE)
#cat(vj,"\n",file = out.file,append = TRUE)
}
} # close if
if(flag.na==1)
{
predict.corr.scca[j.lambda.u:n.lambdas.u, j.lambda.v, j.gamma] <- predict.corr.scca[j.lambda.u:n.lambdas.u, j.lambda.v, j.gamma] + NA
break
}
} # close loop on lambda.u
} # close loop on lambda.v
} # close loop on gamma
} # close cross-validation loop
# ______________Identify optimal sparseness parameter combination___________
predict.corr.scca[is.na(predict.corr.scca)] <- 0
predict.corr.scca <- predict.corr.scca/n.cv
best.predict.corr.scca <- max(abs(predict.corr.scca), na.rm=T)
best.lambda.v <- lambda.v.array[predict.corr.scca==best.predict.corr.scca]
best.lambda.u <- lambda.u.array[predict.corr.scca==best.predict.corr.scca]
best.gamma <- gamma.seq[gamma.array[predict.corr.scca==best.predict.corr.scca]]
cat("best average test sample correlation:",best.predict.corr.scca,"is at",best.lambda.u,"and",best.lambda.v," gamma = ",best.gamma,"\n",file = out.file,append = TRUE)
# ______________________________________________________________________________________________________________
# _____Compute singular vectors using the optimal sparseness parameter combination for the whole data___________
# ______________________________________________________________________________________________________________
cat("begin analyze the whole data:","\n",file = out.file,append = TRUE)
#cat("begin analyze the whole data:","\n",file = "ascca_CV_gamma_iteration.txt",append = TRUE)
k <- sample.sigma12.function(x.data, y.data)
# Get starting values for singular vectors
# as column and row means from matrix K
u.initial <- k %*% ones.q
u.initial <- u.initial /sqrt(as.numeric(t(u.initial)%*%u.initial))
v.initial <- t(k) %*% ones.p
v.initial <- v.initial /sqrt(as.numeric(t(v.initial)%*%v.initial))
uv.svd = svd(k,nu=1,nv=1)
u.svd = uv.svd$u
v.svd = uv.svd$v
uv <- adaptive.scca.function(k, u.initial, v.initial, best.lambda.u, best.lambda.v, u.svd, v.svd, best.gamma)
vj <- uv$v.new # sparse singular vector (canonical vector for Y)
uj <- uv$u.new # sparse singular vector (canonical vector for X)
cat("converged after",uv$i,"iterations.","\n",file = out.file,append = TRUE)
corr.scca <- abs(cor(x.data%*%uj, y.data%*%vj)) # canonical correlation for X and Y data
cat("the final overall correlation is",corr.scca,"\n",file = out.file,append = TRUE)
cat("between",sum(uj != 0),"x variables and",sum(vj != 0),"y variables. \n",file = out.file,append = TRUE)
cat("Transcription Factors \n", file = out.file, append = TRUE)
index.u = uj != 0
t1 = uj[index.u]
t2 = x.map[index.u]
index.map = order(t1,decreasing=T)
index.abs.map = order(abs(t1),decreasing=T)
answer.u = data.frame(t1,t2,sort(t1,decreasing=T),t2[index.map],t1[index.abs.map],t2[index.abs.map])
write.table(answer.u,file = out.file,append = TRUE,row.names = TRUE, col.names = FALSE)
cat("\nTarget Genes \n",file = out.file,append = TRUE)
index.v = vj != 0
t1 = vj[index.v]
t2 = y.map[index.v]
index.map = order(t1,decreasing=T)
index.abs.map = order(abs(t1),decreasing=T)
answer.v = data.frame(t1,t2,sort(t1,decreasing=T),t2[index.map],t1[index.abs.map],t2[index.abs.map])
write.table(answer.v,file = out.file,append = TRUE,row.names = TRUE, col.names = FALSE)