-
Notifications
You must be signed in to change notification settings - Fork 4
/
sim_fisher_p.R
189 lines (139 loc) · 5.04 KB
/
sim_fisher_p.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
##############################################
########### fisher p in simulation ###########
##############################################
# load packages -----
rm(list = ls())
library(tidyverse)
source('~/Trans/plot/theme_my_pub.R')
theme_set(
theme_my_pub(legend.position = "bottom")
)
# I/O & paras -----
file_Sigma <- '/project2/xuanyao/llw/simulation_lambda0.1/new_Sigma/Sigma-new_DGN_module29_K101.rds'
K_all_gene <- 12102
p_cutoff_seq <- c(1e-5, 1e-4, 1e-3, 1e-2, 0.05)
caus.seq <- c(1, 5, 10, 30, 50)/100
var.b <- 0.001
N <- 500
N.sample <- 1
N.sim <- 1e+5
## output -----
file_p_enrich <- str_glue('p_enrich_varb{var.b}_N{N}.rds')
file_p_plot <- str_glue('p_plot_varb{var.b}_N{N}.pdf')
# read files -----
Sigma <- as.matrix(readRDS(file_Sigma))
K <- dim(Sigma)[1]
K_null <- K_all_gene - K
# various caus as models -----
caus.num.seq <- floor(caus.seq * K)
models <- paste0("caus=", caus.seq)
# store enrich p for N.sample snps, various caus as models, and N.sim simulations
p_enrich <- array(
dim = c(N.sample, length(p_cutoff_seq), length(models), N.sim),
dimnames = list(NULL, str_glue('p_{p_cutoff_seq}'), models, NULL)
)
for(i in 1:N.sim){
# simulate effect size b across various caus
B <- matrix(rep(NA, K*length(models)), ncol = length(models),
dimnames = list(NULL, models))
B[, models] <- sapply(caus.num.seq, function(x) as.matrix(c(rnorm(x, sd = sqrt(var.b)), rep(0, K-x))) ) * sqrt(N)
for(model in models){
# Z.alt for z of gene module with caus gene, using Sigma as module correlation
Z.alt <- mvtnorm::rmvnorm(N.sample, B[, model], Sigma)
# Z.null for z of all the other genes with non caus gene, genes are independent
# as it's impossbile to simulation correlation for these non-caus genes
Z.null <- rnorm(K_null*N.sample, 0, 1) %>% matrix(ncol = K_null)
# convert z to p
p.alt <- pchisq(Z.alt^2, df = 1, lower.tail = FALSE)
p.null <- pchisq(Z.null^2, df = 1, lower.tail = FALSE)
# fisher input
aa <- lapply(p_cutoff_seq, function(x) {
rowSums(p.alt < x)
}) %>%
setNames(str_glue('p_{p_cutoff_seq}')) %>%
bind_cols()
bb <- K - aa
cc <- lapply(p_cutoff_seq, function(x) {
rowSums(p.null < x)
}) %>%
setNames(str_glue('p_{p_cutoff_seq}')) %>%
bind_cols()
dd <- K_null - cc
# convert fisher input to hypergeometric test input
# as hypergeometric allows simulataneous calculation of N.sample tests
q = aa
m = aa + cc
n = bb + dd
k = aa + bb
# store hypergeometric test for all N.sample and i-th simulation
p_enrich[, str_glue('p_{p_cutoff_seq}'), model, i] <- lapply(str_glue('p_{p_cutoff_seq}'), function(x) {
phyper(q[[x]], m[[x]], n[[x]], k[[x]], lower.tail = FALSE, log.p = FALSE)
}) %>%
setNames(str_glue('p_{p_cutoff_seq}')) %>%
bind_cols() %>%
unlist()
}
cat("Simulation: ", i, "\n")
if(i %% 1000 == 0) saveRDS(p_enrich, file_p_enrich)
}
# plot of enrich p -----
plt_dat <- as.data.frame.table(p_enrich) %>%
mutate(Var1 = NULL, Var4 = NULL) %>%
rename('p_cutoff' = 'Var2', "model" = 'Var3', 'p_enrich' = 'Freq')
p_upper_show <- 1e-10
plt_dat$p_enrich[plt_dat$p_enrich < p_upper_show] <- p_upper_show
ggpubr::ggarrange(
ggplot(plt_dat) +
geom_boxplot(
aes(x = model, y = -log10(p_enrich), color = p_cutoff),
outlier.alpha = 0.2, outlier.size = 0.1
) +
geom_hline(yintercept = -log10(p_upper_show), linetype = "dashed") +
scale_color_brewer(palette = "Dark2"),
ggplot(plt_dat) +
facet_wrap(~model, nrow = 2) +
geom_freqpoly(
aes(x = -log10(p_enrich), color = p_cutoff),
) +
geom_vline(xintercept = -log10(p_upper_show), linetype = "dashed") +
scale_color_brewer(palette = "Dark2"),
nrow = 2, heights = c(1, 1.5)
)
# save
ggsave(file_p_plot, width = 6, height = 6)
# # qq plot & histogram & power -----
# n_test <- N.sim
#
# # order statistic of null p
# expected <- seq(1, n_test) / (n_test+1)
# lexp <- -log10(expected)
#
# ci_l <- -log10( qbeta(p = (1 - ci_level) / 2, shape1 = 1:n_test, shape2 = n_test:1) )
# ci_r <- -log10( qbeta(p = (1 + ci_level) / 2, shape1 = 1:n_test, shape2 = n_test:1) )
#
# # obs
# observed <- apply(p_enrich, c(2, 3), sort) %>% as.data.frame.table()
# lobs <- -log10(observed)
#
#
# # data for plt
# df_plt = cbind(data.frame(x = lexp, ci_l = ci_l, ci_r = ci_r), lobs) %>%
# pivot_longer(-c(x, ci_l, ci_r), names_to = "Type", values_to = "y")
#
#
# # qq plot
# ggplot(df_plt, aes(x = x, y = y, group = Type)) +
# geom_ribbon(aes(ymin = ci_l, ymax = ci_r), fill = "#e5e5e5", color = "#e5e5e5") +
# geom_abline(slope = 1, intercept = 0, color = "#595959", size = 0.7) +
# geom_point(aes(color = Type), size = 0.5) +
# labs(
# x = bquote(Expected -log[10]~italic((P))),
# y = bquote(Observed -log[10]~italic((P)))
# ) +
# scale_color_manual(
# values = c(
# "#85192d", "#0028a1", "#e89c31",
# RColorBrewer::brewer.pal(8, "Dark2"),
# RColorBrewer::brewer.pal(8, "Set1")
# )
# )