Method verification - simulated gene gain/loss rates
library(panstripe)
library(tidyverse)
library(data.table)
library(ggthemes)
library(lemon)
library(ape)Here, we considered the ability of the panstripe approach to distinguish different rates of gene gain and loss subject to different error rates.
Variable gain/loss rate
set.seed(123)
error_rates <- c(1, 2, 5, 10)
names(error_rates) <- as.character(error_rates)
gl_rates <- c(1e-04, 5e-04, 0.001, 0.01)
nreps <- 5
params <- expand.grid(error_rates, gl_rates, 1:nreps) %>% as_tibble()
colnames(params) <- c("error rate", "gain loss rate", "replicate")
params$sim <- map2(params$`error rate`, params$`gain loss rate`, ~{
simulate_pan(rate = .y, mean_trans_size = 3, fn_error_rate = .x, fp_error_rate = .x,
ngenomes = 100)
})Ability to distinguish gain/loss rates
rates_to_compare <- c(1e-04, 0.001)
params$fit <- map(params$sim, ~{
panstripe(.x$pa, .x$tree, nboot = 0, quiet = TRUE)
})
comparisons_res <- map_dfr(split(params, paste(params$replicate, params$`error rate`)),
function(rep) {
map_dfr(rates_to_compare, function(r1) {
fit1 <- rep$fit[[which(rep$`gain loss rate` == r1)]]
map_dfr(rep$`gain loss rate`, function(r2) {
compare_pangenomes(fit1, rep$fit[[which(rep$`gain loss rate` == r2)]],
nboot = 0)$summary %>% add_column(error_rate = unique(rep$`error rate`)) %>%
add_column(`gain loss rate A` = r1) %>% add_column(`gain loss rate B` = r2) %>%
add_column(replicate = unique(rep$replicate))
})
})
})pdf <- comparisons_res %>% filter(term == "core")
pdf$`gain loss rate B` <- factor(format(pdf$`gain loss rate B`, digits = 3, scientific = TRUE),
levels = format(sort(unique(pdf$`gain loss rate B`)), digits = 3, scientific = TRUE))
pdf$`gain loss rate A` <- format(pdf$`gain loss rate A`, digits = 3, scientific = TRUE)
pdf$error_rate <- factor(paste("Error rate", pdf$error_rate), levels = paste("Error rate",
sort(unique(pdf$error_rate))))
ggplot(pdf, aes(x = `gain loss rate B`, y = -log10(p.value))) + geom_jitter(height = 0,
width = 0.1) + geom_hline(yintercept = -log10(0.05), col = "red") + lemon::facet_rep_grid(`gain loss rate A` ~
error_rate) + theme_clean(base_size = 16) + theme(plot.background = element_blank(),
legend.background = element_blank()) + theme(panel.border = element_blank(),
axis.line = element_line()) + # theme(axis.text.x = element_text(angle = 45, hjust=1)) +
scale_y_sqrt() + xlab("symmetric gene gain/loss rate") + ylab("-log10(p-value)")
ggsave("./figures/gl_rate_comparison.png", width = 10, height = 7)
ggsave("./figures/gl_rate_comparison.pdf", width = 10, height = 7)Ability to distinguish error rates
rates_to_compare <- c(1, 5)
comparisons_err_res <- map_dfr(split(params, paste(params$replicate, params$`gain loss rate`)),
function(rep) {
map_dfr(rates_to_compare, function(r1) {
fit1 <- rep$fit[[which(rep$`error rate` == r1)]]
map_dfr(rep$`error rate`, function(r2) {
compare_pangenomes(fit1, rep$fit[[which(rep$`error rate` == r2)]],
nboot = 0)$summary %>% add_column(`gain loss rate` = unique(rep$`gain loss rate`)) %>%
add_column(`error rate A` = r1) %>% add_column(`error rate B` = r2) %>%
add_column(replicate = unique(rep$replicate))
})
})
})pdf <- comparisons_err_res %>% filter(term == "istip")
pdf$`gain loss rate` <- factor(format(pdf$`gain loss rate`, digits = 3, scientific = TRUE),
levels = format(sort(unique(pdf$`gain loss rate`)), digits = 3, scientific = TRUE))
pdf$`error rate A` <- factor(paste("Error rate", pdf$`error rate A`), levels = paste("Error rate",
sort(unique(pdf$`error rate A`))))
pdf$`error rate B` <- factor(pdf$`error rate B`, levels = sort(unique(pdf$`error rate B`)))
ggplot(pdf, aes(x = `error rate B`, y = -log10(p.value))) + geom_point() + geom_hline(yintercept = -log10(0.05),
col = "red") + lemon::facet_rep_grid(`error rate A` ~ `gain loss rate`) + theme_clean(base_size = 20) +
theme(plot.background = element_blank(), legend.background = element_blank()) +
theme(panel.border = element_blank(), axis.line = element_line()) + scale_y_sqrt() +
xlab("error rate") + ylab("-log10(p-value)")
ggsave("./figures/error_rate_comparison.png", width = 10, height = 7)
ggsave("./figures/error_rate_comparison.pdf", width = 10, height = 7)Ability to distinguish the rate versus size of gene gain/loss events
recomb_size <- c(2, 3, 4, 5)
gl_rates <- c(0.001, 0.01)
nreps <- 5
params <- expand.grid(recomb_size, gl_rates, 1:nreps) %>% as_tibble()
colnames(params) <- c("recomb size", "gain loss rate", "replicate")
params$sim <- map2(params$`recomb size`, params$`gain loss rate`, ~{
simulate_pan(rate = .y, mean_trans_size = .x, fn_error_rate = 2, fp_error_rate = 2,
ngenomes = 200)
})params$fit <- map(params$sim, ~{
panstripe(.x$pa, .x$tree, nboot = 0)
})
rates_to_compare <- c(2)
comparisons_recomb <- map_dfr(split(params, paste(params$replicate, params$`gain loss rate`)),
function(rep) {
map_dfr(rates_to_compare, function(r1) {
fit1 <- rep$fit[[which(rep$`recomb size` == r1)]]
map_dfr(rep$`recomb size`, function(r2) {
compare_pangenomes(fit1, rep$fit[[which(rep$`recomb size` == r2)]],
nboot = 0)$summary %>% add_column(`gain loss rate` = unique(rep$`gain loss rate`)) %>%
add_column(`recomb size A` = r1) %>% add_column(`recomb size B` = r2) %>%
add_column(replicate = unique(rep$replicate))
})
})
})
pdf <- comparisons_recomb %>% filter(term == "dispersion model")
pdf$`gain loss rate` <- factor(format(pdf$`gain loss rate`, digits = 3, scientific = TRUE),
levels = format(sort(unique(pdf$`gain loss rate`)), digits = 3, scientific = TRUE))
pdf$`recomb size A` <- factor(paste("Recombination size", pdf$`recomb size A`), levels = paste("Recombination size",
sort(unique(pdf$`recomb size A`))))
pdf$`recomb size B` <- factor(pdf$`recomb size B`, levels = sort(unique(pdf$`recomb size B`)))
ggplot(pdf, aes(x = `recomb size B`, y = -log10(p.value))) + geom_point() + geom_hline(yintercept = -log10(0.05),
col = "red") + lemon::facet_rep_grid(`recomb size A` ~ `gain loss rate`) + theme_clean(base_size = 20) +
theme(plot.background = element_blank(), legend.background = element_blank()) +
theme(panel.border = element_blank(), axis.line = element_line()) + scale_y_sqrt() +
xlab("recombination size") + ylab("-log10(p-value)")
ggsave("./figures/recombination_size_comparison.png", width = 15, height = 7)
ggsave("./figures/recombination_size_comparison.pdf", width = 15, height = 7)Compare impact of changing gain/loss rate on size estiamtes
recomb_size <- c(2, 3, 4, 5)
gl_rates <- c(1e-04, 5e-04, 0.001, 0.005, 0.01)
nreps <- 5
params <- expand.grid(recomb_size, gl_rates, 1:nreps) %>% as_tibble()
colnames(params) <- c("recomb size", "gain loss rate", "replicate")
params$sim <- map2(params$`recomb size`, params$`gain loss rate`, ~{
simulate_pan(rate = .y, mean_trans_size = .x, fn_error_rate = 2, fp_error_rate = 2,
ngenomes = 200)
})params$fit <- map(params$sim, ~{
panstripe(.x$pa, .x$tree, nboot = 0)
})
rates_to_compare <- c(0.001, 0.01)
comparisons_recomb_same <- map_dfr(split(params, paste(params$replicate, params$`recomb size`)),
function(rep) {
map_dfr(rates_to_compare, function(r1) {
fit1 <- rep$fit[[which(rep$`gain loss rate` == r1)]]
map_dfr(unique(rep$`gain loss rate`)[unique(rep$`gain loss rate`) !=
r1], function(r2) {
compare_pangenomes(fit1, rep$fit[[which(rep$`gain loss rate` == r2)]],
nboot = 0)$summary %>% add_column(`recomb size` = unique(rep$`recomb size`)) %>%
add_column(`gain loss rate A` = r1) %>% add_column(`gain loss rate B` = r2) %>%
add_column(replicate = unique(rep$replicate))
})
})
})
pdf <- comparisons_recomb_same %>% filter(term == "dispersion model")
pdf$`gain loss rate A` <- factor(format(pdf$`gain loss rate A`, digits = 3, scientific = TRUE),
levels = format(sort(unique(pdf$`gain loss rate A`)), digits = 3, scientific = TRUE))
pdf$`gain loss rate B` <- factor(format(pdf$`gain loss rate B`, digits = 3, scientific = TRUE),
levels = format(sort(unique(pdf$`gain loss rate B`)), digits = 3, scientific = TRUE))
pdf$`recomb size` <- factor(paste("Recombination size", pdf$`recomb size`), levels = sort(paste("Recombination size",
unique(pdf$`recomb size`))))
ggplot(pdf, aes(x = `gain loss rate B`, y = -log10(p.value))) + geom_point() + geom_hline(yintercept = -log10(0.05),
col = "red") + lemon::facet_rep_grid(`gain loss rate A` ~ `recomb size`) + theme_clean(base_size = 20) +
theme(plot.background = element_blank(), legend.background = element_blank()) +
theme(panel.border = element_blank(), axis.line = element_line()) + scale_y_sqrt() +
xlab("Gene gain/loss rate") + ylab("-log10(p-value)")
ggsave("./figures/recombination_size_comparison_nochange.png", width = 15, height = 7)
ggsave("./figures/recombination_size_comparison_nochange.pdf", width = 15, height = 7)