ThermalAdaptation/6. Gene whitelist generation using EdgeR at main · AlexFHart/ThermalAdaptation · GitHub

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
### During preliminary analysis we identified a major batch effect between years that was not sufficiently dealt with by batch effect moderators such as limma's RemoveBatchEffect.
### Thus, the following code was used to whitelist genes which were at least somewhat consistent between years for further analysis.
### Based on regression and correlation, you can pick a threshold that's appropriate for your dataset. We ended up with ~7.5k genes on our whitelist.

# setup
# import a file containing the normalised cpms of each year generated by edgeR
setwd("~/Library/Mobile Documents/com~apple~CloudDocs/Documents/PostDoc Folder/Scripts & Code/ThermAdapt/2024/the_plan_II/blacklist prep")
both <- read.csv("both_years.csv")

library(ggplot2)
library(dplyr)

# setting up groups/factors for easier processing
both$Pop <- as.factor(both$Pop)
group_data <- both %>% group_by(Origin, Regime, Assay, Batch, Pop)

group_means <- group_data %>% summarise(mean_value = mean(CALMACG00000000006))

gulp <- group_data %>% summarise_if(is.numeric, mean, na.rm = TRUE)


# setting up regression -> R^2 + corr + p-val for every gene
# cpm (2022) x vs. cpm (2024) y

gulp_2022 <- filter(gulp, Batch == "2022")
gulp_2024 <- filter(gulp, Batch == "2024")

glue <-  inner_join(gulp_2022, gulp_2024, by = c("Origin", "Regime", "Assay", "Pop"))
# anything labelled x is 2022, y is 2024 based on the order from the input file

varlist <- glue[, c("Origin", "Regime", "Assay",  "Pop")]
glue <- subset(glue, select = -c(Batch.x, Batch.y, Origin, Regime, Assay, Pop))
glue <- glue %>% select(order(colnames(glue)))
glue <- cbind(varlist, glue)

# single regression:

test_fit <- lm(CALMACG00000000002.x ~ CALMACG00000000002.y, data = glue)
summ <- summary(test_fit)
corr <- summary(test_fit)$adj.r.squared
pvall <- summary(test_fit)$coef[2,4]

# test loop

test_glue <- glue[,5:10]
test_results <- data.frame()

for (i in seq(from=1, to=6, by=2)) {
  x_2022 <- as.data.frame(test_glue[,i])
  x_2022 <- unlist(x_2022)
  y_2024 <- as.data.frame(test_glue[,i+1])
  y_2024 <- unlist(y_2024)
  loop_fit <- lm(x_2022 ~ y_2024)
  loop_corr <- summary(loop_fit)$adj.r.squared
  loop_pvall <- summary(loop_fit)$coef[2,4]
  output <- c(loop_corr, loop_pvall)
  test_results <- rbind(test_results, output)
  colnames(test_results) <- c("corr", "p-val")
}

# now apply to real dataset

real_results <- data.frame()

for (i in seq(from = 5, to = 25568, by = 2)) {
  x_2022 <- as.data.frame(glue[,i])
  x_2022 <- unlist(x_2022)
  y_2024 <- as.data.frame(glue[,i+1])
  y_2024 <- unlist(y_2024)
  loop_fit <- lm(x_2022 ~ y_2024)
  loop_corr <- summary(loop_fit)$adj.r.squared
  loop_pvall <- summary(loop_fit)$coef[2,4]
  output <- c(loop_corr, loop_pvall)
  real_results <- rbind(real_results, output)
  colnames(real_results) <- c("corr", "p-val")
}

hist(real_results$`corr`)
hist(real_results$`p-val`)

# R2 instead

test_R2 <- cor(test_glue$CALMACG00000000002.x, test_glue$CALMACG00000000002.y)

test_results <- data.frame()

for (i in seq(from=1, to=6, by=2)) {
  x_2022 <- (test_glue[,i])
  x_2022 <- unlist(x_2022)
  y_2024 <- (test_glue[,i+1])
  y_2024 <- unlist(y_2024)
  loop_fit <- cor(x_2022, y_2024)
  test_results <- rbind(test_results, loop_fit)
  colnames(test_results) <- c("R2")
}


# real correlation

real_results <- data.frame()

test_log <- glue[,5:25568]
test_log <- log(test_log)

log_results <- data.frame()

for (i in seq(from = 1, to = 25564, by = 2)) {
  x_2022 <- (test_log[,i])
  x_2022 <- unlist(x_2022)
  y_2024 <- (test_log[,i+1])
  y_2024 <- unlist(y_2024)
  loop_fit <- cor.test(x_2022, y_2024)
  print(loop_fit)
  #log_results <- rbind(log_results, loop_fit)
  #colnames(log_results) <- c("R2")
}


# bonus/extra code snippets which might be helpful:
# isolating groups
# example: cold Brazil

comp_2022 <- which(cpms_2022$Origin == "Brazil" & cpms_2022$Regime == "Cold")
comp_2022 <- cpms_2022[comp_2022,]
comp_2022 <- subset(comp_2022, select = -c(X, Origin, Regime, Assay))
comp_2022 <- t(comp_2022)
avg_2022 <- as.data.frame(rowMeans(comp_2022))

comp_2024 <- which(cpms_2024$Origin == "Brazil" & cpms_2024$Regime == "Cold")
comp_2024 <- cpms_2024[comp_2024,]
comp_2024 <- subset(comp_2024, select = -c(X, Origin, Regime, Assay))
comp_2024 <- t(comp_2024)
avg_2024 <- as.data.frame(rowMeans(comp_2024))

# scatter

i_hate_ggplot <- avg_2022
i_hate_ggplot <- as.data.frame(append(i_hate_ggplot, avg_2024))
colnames(i_hate_ggplot) <- c("y2022", "y2024")

library(ggplot2)

ggplotRegression <- function(fit){

  require(ggplot2)

  ggplot(fit$model, aes_string(x = names(fit$model)[2], y = names(fit$model)[1])) +
    geom_point() +
    stat_smooth(method = "lm", col = "red") +
    labs(title = paste("Adj R2 = ",signif(summary(fit)$adj.r.squared, 5),
                       "Intercept =",signif(fit$coef[[1]],5 ),
                       " Slope =",signif(fit$coef[[2]], 5),
                       " P =",signif(summary(fit)$coef[2,4], 5)))
}

ggplotRegression(lm(CALMACG00000000002.x ~ CALMACG00000000002.y, data = test_glue))