-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathS1_2_4_CLPM in ABCD_gene.R
283 lines (220 loc) · 11.1 KB
/
S1_2_4_CLPM in ABCD_gene.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
# %% Step 1.3: CLPM between Sleep and ADHD in ABCD
# P-value was tested by a multi-level block permutation (family relatedness based on genetic data)
# written by Dr Qiang Luo and modifiedy by Chun Shen
# Email: [email protected]
# released on 21 Mar 2020
# please cite: Shen, et al. Biological Psychiatry 2020
# According to the data access policy of the ABCD study, the ABCD data needs to be accessed through NDA.
# Data files used in this code:
# Line 18: data3036ABCD.csv for the behaviour measurements
# Line 181: Pset3036.csv for the multi-level block-permutation generated by
library(lavaan)
library(ggplot2)
library(ppcor)
library(xlsx)
library(parallel)
# load data
data <- read.csv(file="data3036ABCD.csv", header=T) # need the data file for ABCD [excluding those who without genetic data]
data[data==-999] <- NA
# extract data and rename the columns
data.clpm <- data[,c(1:2, 3,6, 4,7,5,8, 17, 9,13, 10,14, 11,15, 12,16, 18:37, 38:40, 41,44, 42,45, 43,46)]
colnames(data.clpm) <- c("adhd1","adhd2",
"para1","para2",
"dyss1","dyss2",
"tot1","tot2",
"sex",
"bmi1","bmi2",
"pub1", "pub2",
"ses11","ses21",
"ses12","ses22",
"s1","s2","s3","s4","s5","s6","s7","s8","s9","s10",
"s11","s12","s13","s14","s15","s16","s17","s18","s19","s20",
"race1", "race2", "race3",
"tr12","tr22",
"tr13","tr23",
"tr14","tr24") # x ~ adhd; y ~ sleep; sex; bmi; pub; ses(dummy 2); site(dummy 20); race; tr(dummy 3);
print(colnames(data)[c(1:2, 3,6, 4,7,5,8, 17, 9,13, 10,14, 11,15, 12,16, 18:37, 38:40, 41,44, 42,45, 43,46)])
# covariates at the baseline: bmi1,pub1,ses11,ses12 sex,s1-s20,race1,race2,race3,tr1_2,tr1_3,tr1_4
# covariates at the followup: bmi2,pub2,ses21,ses22 sex,s1-s20,race1,race2,race3,tr2_2,tr2_3,tr2_4
##########################
# Model 1: cross-lagged path model
##########################
## model 1.1 unconstrained
clpmModel <-
'
#Note, the data contain x1-2 and y1-2
#Latent mean Structure with intercepts
kappa =~ 1*x1 + 1*x2
omega =~ 1*y1 + 1*y2
x1 ~ mu1*1 #intercepts
x2 ~ mu2*1
y1 ~ pi1*1
y2 ~ pi2*1
kappa ~~ 0*kappa #variance
omega ~~ 0*omega #variance
kappa ~~ 0*omega #covariance
#laten vars for AR and cross-lagged effects
p1 =~ 1*x1 #each factor loading set to 1
p2 =~ 1*x2
q1 =~ 1*y1
q2 =~ 1*y2
#regressions
p2 ~ alpha2*p1 + beta2*q1 + sex + bmi2 + pub2 + ses21 + ses22 + s1+s2+s3+s4+s5+s6+s7+s8+s9+s10+s11+s12+s13+s14+s15+s16+s17+s18+s19+s20 + race1+race2+race3 + tr22+tr23+tr24
q2 ~ delta2*q1 + gamma2*p1+ sex + bmi2 + pub2 + ses21 + ses22 + s1+s2+s3+s4+s5+s6+s7+s8+s9+s10+s11+s12+s13+s14+s15+s16+s17+s18+s19+s20 + race1+race2+race3 + tr22+tr23+tr24
p1 ~ sex + bmi1 + pub1 + ses11 + ses12 + s1+s2+s3+s4+s5+s6+s7+s8+s9+s10+s11+s12+s13+s14+s15+s16+s17+s18+s19+s20 + race1+race2+race3 + tr12+tr13+tr14
q1 ~ sex + bmi1 + pub1 + ses11 + ses12 + s1+s2+s3+s4+s5+s6+s7+s8+s9+s10+s11+s12+s13+s14+s15+s16+s17+s18+s19+s20 + race1+race2+race3 + tr12+tr13+tr14
p1 ~~ p1 #variance
p2 ~~ u2*p2
q1 ~~ q1 #variance
q2 ~~ v2*q2
p1 ~~ q1 #p1 and q1 covariance
p2 ~~ q2'
colnames.original <- colnames(data.clpm)
colnames(data.clpm) <- colnames.original
colnames(data.clpm)[c(1,2,7,8)] <- c("x1", "x2", "y1", "y2") # x -- adhd; y -- tot
fit.clpmModel.tot <- lavaan(clpmModel, data = data.clpm,
missing = 'ML', #for the missing data!
int.ov.free = F,
int.lv.free = F,
auto.fix.first = F,
auto.fix.single = F,
auto.cov.lv.x = F,
auto.cov.y = F,
auto.var = F)
colnames(data.clpm) <- colnames.original
colnames(data.clpm)[c(1,2,5,6)] <- c("x1", "x2", "y1", "y2") # x -- adhd; y -- dyss
fit.clpmModel.dyss <- lavaan(clpmModel, data = data.clpm,
missing = 'ML', #for the missing data!
int.ov.free = F,
int.lv.free = F,
auto.fix.first = F,
auto.fix.single = F,
auto.cov.lv.x = F,
auto.cov.y = F,
auto.var = F)
colnames(data.clpm) <- colnames.original
colnames(data.clpm)[c(1,2,3,4)] <- c("x1", "x2", "y1", "y2") # x -- adhd; y -- para
fit.clpmModel.para <- lavaan(clpmModel, data = data.clpm,
missing = 'ML', #for the missing data!
int.ov.free = F,
int.lv.free = F,
auto.fix.first = F,
auto.fix.single = F,
auto.cov.lv.x = F,
auto.cov.y = F,
auto.var = F)
tot.beta2.z <- 0
tot.gamma2.z <- 0
dyss.beta2.z <- 0
dyss.gamma2.z <- 0
para.beta2.z <- 0
para.gamma2.z <- 0
para.all <- data.frame(tot.beta2.z, tot.gamma2.z, dyss.beta2.z, dyss.gamma2.z, para.beta2.z, para.gamma2.z)
result.tot <- summary(fit.clpmModel.tot, standardized = T, fit.measures = TRUE)
#para.all$tot.beta2 <- result.tot$PE[which(result.tot$PE$label=='beta2'),"std.all"]
#para.all$tot.gamma2 <- result.tot$PE[which(result.tot$PE$label=='gamma2'),"std.all"]
para.all$tot.beta2.z <- result.tot$PE[which(result.tot$PE$label=='beta2'),"z"]
para.all$tot.gamma2.z <- result.tot$PE[which(result.tot$PE$label=='gamma2'),"z"]
result.dyss <- summary(fit.clpmModel.dyss, standardized = T, fit.measures = TRUE)
#para.all$dyss.beta2 <- result.dyss$PE[which(result.dyss$PE$label=='beta2'),"std.all"]
#para.all$dyss.gamma2 <- result.dyss$PE[which(result.dyss$PE$label=='gamma2'),"std.all"]
para.all$dyss.beta2.z <- result.dyss$PE[which(result.dyss$PE$label=='beta2'),"z"]
para.all$dyss.gamma2.z <- result.dyss$PE[which(result.dyss$PE$label=='gamma2'),"z"]
result.para <- summary(fit.clpmModel.para, standardized = T, fit.measures = TRUE)
#para.all$para.beta2 <- result.para$PE[which(result.para$PE$label=='beta2'),"std.all"]
#para.all$para.gamma2 <- result.para$PE[which(result.para$PE$label=='gamma2'),"std.all"]
para.all$para.beta2.z <- result.para$PE[which(result.para$PE$label=='beta2'),"z"]
para.all$para.gamma2.z <- result.para$PE[which(result.para$PE$label=='gamma2'),"z"]
print(para.all)
colnames(data.clpm) <- colnames.original
#p.fdr <- as.data.frame(p.adjust(result$PE[c(40:59),]$pvalue, method = "BH"))
#rownames(p.fdr) <- result$PE[c(40:59),]$label
#colnames(p.fdr) <- 'p.fdr'
#print(p.fdr)
##################
## permutation
##################
permID <- read.csv("Pset3036.csv", header = T)
cl = makeCluster(12)
# check it works.
clusterEvalQ(cl, runif(10)) # -> this works
clusterSetRNGStream(cl, 100) # to get reproduciable results
rm(results)
nperm <- 5000
processInput <- function(i){
# permute
data.clpm.perm <- data.clpm
data.clpm.perm[,c(1,2)] <- data.clpm.perm[permID[,i],c(1,2)] # permute ADHD score at the baseline
# model for dyss
colnames(data.clpm.perm) <- colnames.original
colnames(data.clpm.perm)[c(1,2,5,6)] <- c("x1", "x2", "y1", "y2") # x -- adhd; y -- dyss
fit.clpmModel.dyss.perm <- lavaan(clpmModel, data = data.clpm.perm,
missing = 'ML', #for the missing data!
int.ov.free = F,
int.lv.free = F,
auto.fix.first = F,
auto.fix.single = F,
auto.cov.lv.x = F,
auto.cov.y = F,
auto.var = F)
result.dyss <- summary(fit.clpmModel.dyss.perm, standardized = T, fit.measures = TRUE)
dyss.beta2.z <- result.dyss$PE[which(result.dyss$PE$label=='beta2'),"z"]
dyss.gamma2.z <- result.dyss$PE[which(result.dyss$PE$label=='gamma2'),"z"]
# model for para
colnames(data.clpm.perm) <- colnames.original
colnames(data.clpm.perm)[c(1,2,3,4)] <- c("x1", "x2", "y1", "y2") # x -- adhd; y -- para
fit.clpmModel.para.perm <- lavaan(clpmModel, data = data.clpm.perm,
missing = 'ML', #for the missing data!
int.ov.free = F,
int.lv.free = F,
auto.fix.first = F,
auto.fix.single = F,
auto.cov.lv.x = F,
auto.cov.y = F,
auto.var = F)
result.para <- summary(fit.clpmModel.para.perm, standardized = T, fit.measures = TRUE)
para.beta2.z <- result.para$PE[which(result.para$PE$label=='beta2'),"z"]
para.gamma2.z <- result.para$PE[which(result.para$PE$label=='gamma2'),"z"]
# model for tot
colnames(data.clpm.perm) <- colnames.original
colnames(data.clpm.perm)[c(1,2,7,8)] <- c("x1", "x2", "y1", "y2") # x -- adhd; y -- tot
fit.clpmModel.tot.perm <- lavaan(clpmModel, data = data.clpm.perm,
missing = 'ML', #for the missing data!
int.ov.free = F,
int.lv.free = F,
auto.fix.first = F,
auto.fix.single = F,
auto.cov.lv.x = F,
auto.cov.y = F,
auto.var = F)
result.tot <- summary(fit.clpmModel.tot.perm, standardized = T, fit.measures = TRUE)
tot.beta2.z <- result.tot$PE[which(result.tot$PE$label=='beta2'),"z"]
tot.gamma2.z <- result.tot$PE[which(result.tot$PE$label=='gamma2'),"z"]
para.perm.all <- data.frame(tot.beta2.z, tot.gamma2.z, dyss.beta2.z, dyss.gamma2.z, para.beta2.z, para.gamma2.z)
return(para.perm.all)
}
clusterEvalQ(cl, library(lavaan)) # load the library to every core
clusterExport(cl, c("processInput","data.clpm", "permID", "colnames.original", "clpmModel")) # export the variables to every core
results <- parLapply(cl, 1:5000, function(i) processInput(i))
#results <- list()
#results[[1]] <- results.1
#t <- 1
#data.temp <- results.5000
#for (i in 1:length(data.temp)){
# t <- t + 1
# results[[t]] <- data.temp[[i]]
#}
para.perm.all <- data.frame(matrix(0,nrow = nperm, ncol = 6))
colnames(para.perm.all) <- colnames(para.all)
for (i in c(1:nperm)){
para.perm.all[i,] <- as.data.frame(results[[i]])
}
rm(results)
max.tot <- matrix(apply(abs(para.perm.all[,c(1,2)]), 1, function(x){ max(x) }), nrow = nperm, ncol = 1)
max.dimension <- matrix(apply(abs(para.perm.all[,c(3,6)]), 1, function(x){ max(x) }), nrow = nperm, ncol = 1)
perm.p <- matrix(1, nrow = 1, ncol = 6)
perm.p[1,c(1,2)] <- apply(abs(para.all[c(1:2)]), 2, function(x) {sum(max.tot > x) / nperm})
perm.p[1,c(3:6)] <- apply(abs(para.all[c(3:6)]), 2, function(x) {sum(max.dimension > x) / nperm})
colnames(perm.p) <- colnames(para.all)
print(perm.p)
stopCluster(cl)