-
Notifications
You must be signed in to change notification settings - Fork 0
/
real_data_example.R
183 lines (169 loc) · 7.43 KB
/
real_data_example.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
# Run dependent_sim on a real RNA-seq dataset
library(dependentsimr)
library(ggplot2)
library(ggpointdensity)
library(tidyr)
library(dplyr)
# Load the real data ---------------------------------
# from GEO: GSE77221
head(Weger18)
weger18_library_sizes <- apply(Weger18, 2, sum)
# Run dependent_sim on this data --------------------
N_SAMPLES <- 96
library_sizes <- (weger18_library_sizes + rep(0, N_SAMPLES)) / mean(weger18_library_sizes) # Recycle the existing library sizes
rs <- get_random_structure(list(counts=Weger18), method="pca", rank=2, type="DESeq2")
draws <- draw_from_multivariate_corr(rs, n_samples=N_SAMPLES, size_factors=library_sizes)$counts
rs_corpcor <- get_random_structure(list(counts=Weger18), method="corpcor", type="DESeq2")
draws_corpcor <- draw_from_multivariate_corr(rs_corpcor, n_samples=N_SAMPLES, size_factors=library_sizes)$counts
rs_wishart <- get_random_structure(list(counts=Weger18), method="spiked Wishart", type="DESeq2")
draws_wishart <- draw_from_multivariate_corr(rs_wishart, n_samples=N_SAMPLES, size_factors=library_sizes)$counts
# Generate simulated data without any dependence
rs_indep <- remove_dependence(rs)
indep_draws <- draw_from_multivariate_corr(rs_indep, n_samples=N_SAMPLES, size_factors=library_sizes)$counts
# Scale to Counts Per Million ------------------------
cpm <- function(x) { # counts per million
return(t(t(x) / apply(x, 2, sum) * 1e6))
}
scaled_read_data <- cpm(Weger18)
scaled_draws <- cpm(draws)
scaled_draws_wishart <- cpm(draws_wishart)
scaled_draws_corpcor <- cpm(draws_corpcor)
scaled_indep_draws <- cpm(indep_draws)
# Plot marginal distributions --------------------------
simulated_mean <- apply(scaled_draws[,1:ncol(Weger18)], 1, mean)
simulated_variance <- apply(scaled_draws[,1:ncol(Weger18)], 1, var)
real_mean <- apply(scaled_read_data, 1, mean)
real_variance <- apply(scaled_read_data, 1, var)
marg_dist <- data.frame(
real_mean = real_mean,
simulated_mean = simulated_mean,
real_variance = real_variance,
simulated_variance = simulated_variance
)
ggplot(marg_dist|> filter(real_mean > 0.1))+
geom_pointdensity(aes(x=log(real_mean+1),y=log(simulated_mean+1))) +
geom_abline(slope=1, intercept=0)
ggplot(marg_dist|> filter(real_mean > 0.1))+
geom_pointdensity(aes(x=log(real_variance+1),y=log(simulated_variance+1))) +
geom_abline(slope=1, intercept=0)
# Plot gene-gene correlation ---------------------------
N_rows <- 3000
high_expr_rows <- which(apply(Weger18, 1, mean) > 300)
selected_rows <- sample(high_expr_rows, N_rows)
selected_samples <- sample(ncol(draws), ncol(Weger18))
real_corr <- cor(t(scaled_read_data[selected_rows,]))
sim_corr <- cor(t(scaled_draws[selected_rows,selected_samples]))
indep_sim_corr <- cor(t(scaled_indep_draws[selected_rows,selected_samples]))
probs <- (1:1000)/1000
real_corr_q <- quantile(real_corr, probs = probs, na.rm=TRUE)
sim_corr_q <- quantile(sim_corr, probs = probs, na.rm=TRUE)
indep_sim_corr_q <- quantile(indep_sim_corr, probs = probs, na.rm=TRUE)
ggplot(data = rbind(data.frame(
real_corr_q = real_corr_q,
sim_corr_q = sim_corr_q,
type="dependent"
),
data.frame(
real_corr_q = real_corr_q,
sim_corr_q = indep_sim_corr_q,
type = "independent"
))) +
geom_point(aes(x = real_corr_q, y = sim_corr_q, color=type)) +
geom_abline(slope=1, intercept=0) +
labs(
x = "Correlation coefficient (real data)",
y = "Correlation coefficient (simulated data)",
title = "QQ of Simulated Correlation",
subtitle = paste("In", N_rows, "randomly chosen genes"))
# Plot the top principal components --------------------
constant_rows <- apply(Weger18, 1, function(x) all(x - mean(x) == 0))
pca <- prcomp(t(scaled_read_data[!constant_rows,]), rank.=2, scale.=TRUE)
projected_read_data <- predict(pca, t(scaled_read_data[!constant_rows, ]))
projected_draws <- predict(pca, t(scaled_draws[!constant_rows, ]))
projected_indep_draws <- predict(pca, t(scaled_indep_draws[!constant_rows, ]))
both_data <- data.frame(list(
PC1 = c(projected_read_data[,"PC1"], projected_draws[,"PC1"], projected_indep_draws[,"PC1"]),
PC2 = c(projected_read_data[,"PC2"], projected_draws[,"PC2"], projected_indep_draws[,"PC2"]),
type = c(
rep("real", nrow(projected_read_data)),
rep("sim", nrow(projected_draws)),
rep("indep_sim", nrow(projected_indep_draws))
)
))
both_data$type <- factor(both_data$type, levels=c("real", "sim", "indep_sim"))
ggplot(data = both_data, aes(x=PC1, y=PC2,color=type)) +
geom_point() +
labs(
title = "Top 2 PCA Components of the Real Data",
subtitle = "Projections onto the PCA components of the real data"
) +
scale_color_manual(values = list(real="blue", sim="#00BFC4", indep_sim="#F8766D"))
# Compare SVD with and without dependence ----------------
PCA <- function(data) {
constant_rows <- apply(data, 1, function(x) all(x - mean(x) == 0))
return(prcomp(t(data[!constant_rows,]), rank.=2, scale.=TRUE))
}
pca_sdevs <- tibble(
type = "real",
component = 1:12,
sdev = PCA(scaled_read_data)$sdev,
)
print("Running PCAs")
for (i in 0:7) {
pca_sdevs <- rbind(
pca_sdevs,
tibble(
type = "sim dep",
component = 1:12,
sdev = PCA(scaled_draws[,(1+12*i):(12*(i+1))])$sdev,
),
tibble(
type = "sim indep",
component = 1:12,
sdev = PCA(scaled_indep_draws[,(1+12*i):(12*(i+1))])$sdev,
),
deparse.level=1
)
}
ggplot(data=pca_sdevs |> filter(component<6)) +
facet_grid(cols=vars(component)) +
geom_boxplot(aes(x=type, y=sdev)) +
labs(title="Size of top PCA components")
# Run DESeq2 on it ---------------------------------------
# Do a 5v5 study. We expect no significant differences since all are drawn from the same samples
low_disp_rs <- rs
low_disp_rs$marginals$counts$dispersion <- rs$marginals$counts$dispersion/1
low_disp_draws <- draw_from_multivariate_corr(low_disp_rs, n_samples=30, size_factors=library_sizes[1:30])$counts
dds <- DESeq2::DESeqDataSetFromMatrix(draws[,1:12], tibble(group = rep(c("ctrl", "case"), 6)), ~group)
dds <- DESeq2::DESeq(dds)
res <- DESeq2::results(dds)
ggplot(tibble(pvalue=res$pvalue), aes(sample=pvalue)) + stat_qq(distribution=stats::qunif) + geom_abline(slope=1, intercept=0) + scale_x_log10() + scale_y_log10()
ggplot(tibble(
deseq_dispersion = DESeq2::dispersions(dds),
gene_dispersion = S4Vectors::mcols(dds)$dispGeneEst,
true_dispersion = rs$marginals$counts$dispersion
)) + geom_pointdensity(aes(true_dispersion, gene_dispersion)) +
scale_x_log10() +
scale_y_log10() +
geom_abline(slope=1, intercept=0)
# Other comparisons of real to simulated data ------------
long_marg_dist <- marg_dist |>
pivot_longer(cols=everything(), names_to=c("source", "var"), names_pattern="(.*)_(.*)", values_to="value")
ggplot(long_marg_dist |> filter(var == "variance")) +
geom_freqpoly(aes(log10(value), color=source)) +
labs(x="log10(variance)")
ggplot(long_marg_dist |> filter(var == "mean")) +
geom_freqpoly(aes(log10(value), color=source)) +
labs(x="log10(mean)")
ggplot(marg_dist, aes(x=log10(real_mean), y=log10(real_variance))) + geom_point()
ggplot(tibble(
deseq_dispersion = DESeq2::dispersions(dds),
est_dispersion = S4Vectors::mcols(dds)$dispGeneEst,
simulated_dispersion = rs$marginals$counts$dispersion,
simulated_mean = rs$marginals$counts$q,
deseq_mean = res$baseMean
), aes(x=deseq_mean, y=deseq_dispersion)) +
geom_point(aes(x=simulated_mean, y=simulated_dispersion), color="blue", size=3) +
geom_point(aes(x=deseq_mean, y=est_dispersion), color="red", size=2) +
geom_point(size=1) +
scale_x_log10() + scale_y_log10()