xgx theme
The xgx_theme() function includes the xGx recommended plot settings. It sets the background to white with light grey lines for the major and minor breaks. This minimizes chart ink as recommended by Edward Tufte. You can add xgx_theme() to an existing ggplot object, or you can call xgx_plot() in place of ggplot() for all of your plot initiations.
xgx_plot(mtcars, aes(x = cyl, y = mpg)) + geom_point()
You may wish to set the theme to xgx_theme for your R session, as we do below.
theme_set(xgx_theme())
## Alternative, equivalent function:
xgx_theme_set()
# time <- rep(seq(1,10),5)
# id <- sort(rep(seq(1,5), 10))
# conc <- exp(-time)*sort(rep(rlnorm(5),10))
#
# data <- data.frame(time = time, concentration = conc, id = factor(id))
# xgx_plot() + xgx_geom_spaghetti(data = data, mapping = aes(x = time, y = concentration, group = id, color = id))
#
# xgx_spaghetti(data = data, mapping = aes(x = time, y = concentration, group = id, color = id))
Confidence intervals
The code for confidence intervals is a bit complex and hard to remember. Rather than copy-pasting this code we provide the function xgx_stat_ci for calculating and plotting default confidence intervals and also xgx_geom_ci for percentile intervals. xgx_stat_ci allows the definition of multiple geom options in one function call, defined through a list. The default is geom = list("point","line","errorbar"). Additional ggplot options can be fed through the ggplot object call, or the xgx_stat_ci layer. (Note that xgx_stat_ci and xgx_geom_ci are equivalent). xgx_stat_pi and xgx_geom_pi work in a similar fashion but for percentile intervals.
data <- data.frame(x = rep(c(1, 2, 3), each = 20),
y = rep(c(1, 2, 3), each = 20) + stats::rnorm(60),
group = rep(1:3, 20))
xgx_plot(data,aes(x = x, y = y)) +
xgx_stat_ci(conf_level = .95)
xgx_plot(data,aes(x = x, y = y)) +
xgx_stat_pi(percent = .95)
xgx_plot(data,aes(x = x, y = y)) +
xgx_stat_ci(conf_level = .95, geom = list("pointrange","line"))
xgx_plot(data,aes(x = x, y = y)) +
xgx_stat_ci(conf_level = .95, geom = list("ribbon","line"))
xgx_plot(data,aes(x = x, y = y, group = group, color = factor(group))) +
xgx_stat_ci(conf_level = .95, alpha = 0.5,
position = position_dodge(width = 0.5))
The default settings calculate the confidence interval based on the Student t Distribution (assuming normally distributed data). You can also specify “lognormal”" or “binomial”" for the distribution. The former will perform the confidence interval operation on the log-scaled data, the latter uses the binomial exact confidence interval calculation from the binom package.
Note: you DO NOT need to use both distribution = "lognormal" and scale_y_log10(), choose only one of these.
# plotting lognormally distributed data
data <- data.frame(x = rep(c(1, 2, 3), each = 20),
y = 10^(rep(c(1, 2, 3), each = 20) + stats::rnorm(60)),
group = rep(1:3, 20))
xgx_plot(data, aes(x = x, y = y)) +
xgx_stat_ci(conf_level = 0.95, distribution = "lognormal")
# note: you DO NOT need to use both distribution = "lognormal" and scale_y_log10()
xgx_plot(data,aes(x = x, y = y)) +
xgx_stat_ci(conf_level = 0.95) + xgx_scale_y_log10()
# plotting binomial data
data <- data.frame(x = rep(c(1, 2, 3), each = 20),
y = rbinom(60, 1, rep(c(0.2, 0.6, 0.8), each = 20)),
group = rep(1:3, 20))
xgx_plot(data, aes(x = x, y = y)) +
xgx_stat_ci(conf_level = 0.95, distribution = "binomial")
Nice log scale
This version of the log scale function shows the tick marks between the major breaks (i.e. at 1, 2, 3, … 10, instead of just 1 and 10). It also uses \[10^x\] notation when the labels are base 10 and are very small or very large (<.001 or >9999)
df <- data.frame(x = c(0, stats::rlnorm(1000, 0, 1)),
y = c(0, stats::rlnorm(1000, 0, 3)))
xgx_plot(data = df, aes(x = x, y = y)) +
geom_point() +
xgx_scale_x_log10() +
xgx_scale_y_log10()
#> Warning: Transformation introduced infinite values in continuous x-axis
#> Warning: Transformation introduced infinite values in continuous y-axis
Nice scale for percent change data
This transform is useful for plotting percent change from baseline data. Percent change data can range from -100% to +Inf%, and depending on the range of the data, a linear scale can lose the desired resolution. This transform plots percent change data on a scale of log10(PCHG + 100%), similar to a log scale of ratio to baseline.
Nsubj <- 10
Doses <- c(0, 25, 50, 100, 200)
Ntot <- Nsubj*length(Doses)
times <- c(0,14,30,60,90)
dat1 <- data.frame(ID = 1:(Ntot),
DOSE = rep(Doses, Nsubj),
PD0 = rlnorm(Ntot, log(100), 1),
Kout = exp(rnorm(Ntot,-2, 0.3)),
Imax = 1,
ED50 = 25) %>%
dplyr::mutate(PDSS = PD0*(1 - Imax*DOSE/(DOSE + ED50))*exp(rnorm(Ntot, 0.05, 0.3)) ) %>%
merge(data.frame(ID = rep(1:(Ntot), each = length(times)), Time = times), by = "ID") %>%
dplyr::mutate(PD = ((PD0 - PDSS)*(exp(-Kout*Time)) + PDSS),
PCHG = (PD - PD0)/PD0)
ggplot2::ggplot(dat1 %>% subset(Time == 90),
ggplot2::aes(x = DOSE, y = PCHG, group = DOSE)) +
ggplot2::geom_boxplot() +
xgx_theme() +
xgx_scale_y_percentchangelog10() +
ylab("Percent Change from Baseline") +
xlab("Dose (mg)")
ggplot2::ggplot(dat1,
ggplot2::aes(x = Time, y = PCHG, group = ID, color = factor(DOSE))) +
ggplot2::geom_line() +
xgx_theme() +
xgx_scale_y_percentchangelog10() +
guides(color = guide_legend(title = "Dose (mg)")) +
ylab("Percent Change from Baseline")
dat2 <- data.frame(ID = 1:(Ntot),
DOSE = rep(Doses, Nsubj),
PD0 = rlnorm(Ntot, log(100), 1),
Kout = exp(rnorm(Ntot,-2, 0.3)),
Emax = 50*rlnorm(Ntot, 0, 0.3),
ED50 = 300) %>%
dplyr::mutate(PDSS = PD0*(1 + Emax*DOSE/(DOSE + ED50))*exp(rnorm(Ntot, -1, 0.3)) ) %>%
merge(data.frame(ID = rep(1:(Ntot), each = length(times)), Time = times), by = "ID") %>%
dplyr::mutate(PD = ((PD0 - PDSS)*(exp(-Kout*Time)) + PDSS),
PCHG = (PD - PD0)/PD0)
ggplot2::ggplot(dat2, ggplot2::aes(x = DOSE, y = PCHG, group = DOSE)) +
ggplot2::geom_boxplot() +
xgx_theme() +
xgx_scale_y_percentchangelog10() +
ylab("Percent Change from Baseline") +
xlab("Dose (mg)")
ggplot2::ggplot(dat2,
ggplot2::aes(x = Time, y = PCHG, group = ID, color = factor(DOSE))) +
ggplot2::geom_line() +
xgx_theme() +
xgx_scale_y_percentchangelog10() +
guides(color = guide_legend(title = "Dose (mg)")) +
ylab("Percent Change from Baseline")
Scaling x-axis as a time scale
For time, it’s often good for the x ticks to be spaced in a particular way. For instance, for hours, subdividing in increments by 24, 12, 6, and 3 hours can make more sense than by 10 or 100. Similarly for days, increments of 7 or 28 days are preferred over 5 or 10 days. xgx_scale_x_time_units allows for this, where it is the input and output units.
data <- data.frame(x = 1:1000, y = stats::rnorm(1000))
g <- xgx_plot(data = data, aes(x = x, y = y)) +
geom_point()
g1 <- g + xgx_scale_x_time_units(units_dataset = "hours", units_plot = "hours")
g2 <- g + xgx_scale_x_time_units(units_dataset = "hours", units_plot = "days")
g3 <- g + xgx_scale_x_time_units(units_dataset = "hours", units_plot = "weeks")
g4 <- g + xgx_scale_x_time_units(units_dataset = "hours", units_plot = "months")
gridExtra::grid.arrange(g1, g2, g3, g4, nrow = 2)
