绘图散点图中的颜色编码误差条

Question

I'm trying to create a forest plot with R plotly where I want to color code the effect sizes (points) and their error bars by their corresponding p-values.

Here are toy data:

set.seed(1)

factors <- paste0(1:25,":age")
effect.sizes <- rnorm(25,0,1)
effect.errors <- abs(rnorm(25,0,1))
p.values <- runif(25,0,1)

Here's what I'm trying:

library(dplyr)
plotly::plot_ly(type='scatter',mode="markers",y=~factors,x=~effect.sizes,color=~p.values,colors=grDevices::colorRamp(c("darkred","gray"))) %>%
      plotly::add_trace(error_x=list(array=effect.errors),marker=list(color=~p.values,colors=grDevices::colorRamp(c("darkred","gray")))) %>%
      plotly::colorbar(limits=c(0,1),len=0.4,title="P-Value") %>%
      plotly::layout(xaxis=list(title="Effect Size",zeroline=T,showticklabels=T),yaxis=list(title="Factor",zeroline=F,showticklabels=T))

which gives me:

Which is pretty close to what I want except for:

1. I'd like the error bars to be colored similar to the effect sizes (by the corresponding p-values).
2. Remove the two trace legends below the colorbar
3. Have the order of the labels on the y-axis be that of factors

Any idea?

Answer 1

Okay it took me a while to warm up my plotly skills. Since your first point was the most difficult, I will go reversely through your points.

• 3. That can be achied by manipulating the layout using categoryorder and categoryarray in the yaxis -list (cf. motos answer here )
• 2. Set showlegend=FALSE
• 1. That was tricky. I had to move your second line (the error bars) in the first. Added a color vector to it. Put it in the plot_ly -function. Used split to allow the correct coloring by group. Added the color for the points in a marker -list. In additon I converted the p.values via the colorRamp to hex -because every simpler solution didn't work for me.

Looks like this:

The code (the colorbar created some issues):

### Set category order
yform <- list(categoryorder = "array",
              categoryarray = rev(factors),
              title="Factor",zeroline=F,showticklabels=T)

### set the color scale and convert it to hex
library(grDevices)
mycramp<-colorRamp(c("darkred","gray"))
mycolors<-rgb(mycramp(p.values),maxColorValue = 255)

### plot without the adjusted colorbar
library(plotly)
### Without colorbar adjustment
  plot_ly(type='scatter',mode="markers",y=~factors,x=~effect.sizes,
          color=~p.values,colors=grDevices::colorRamp(c("darkred","gray")),
          error_x=list(array=effect.errors,color=mycolors),split=factors,showlegend=FALSE,marker=list(color=mycolors)) %>%
      layout(xaxis=list(title="Effect Size",zeroline=T,showticklabels=T),yaxis=yform)

  ### The colorbar-adjustment kicks out the original colors of the scatter points. Either you plot them over
  plot_ly(type='scatter',mode="markers",y=~factors,x=~effect.sizes,
          color=~p.values,colors=grDevices::colorRamp(c("darkred","gray")),
          error_x=list(array=effect.errors,color=mycolors),split=factors,showlegend=FALSE,marker=list(color=mycolors)) %>%
      layout(xaxis=list(title="Effect Size",zeroline=T,showticklabels=T),yaxis=yform) %>%
  colorbar(limits=c(0,1),len=0.4,title="P-Value",inherit=FALSE) %>%
      add_trace(type='scatter',mode="markers",y=~factors,x=~effect.sizes,
            showlegend=FALSE,marker=list(color=mycolors),inherit=FALSE) %>%
    layout(xaxis=list(title="Effect Size",zeroline=T,showticklabels=T),yaxis=yform)

  ### or you try to set the colorbar before the plot. This results in some warnings
  plot_ly() %>%
  colorbar(limits=c(0,1),len=0.4,title="P-Value",inherit=FALSE) %>%
      add_trace(type='scatter',mode="markers",y=~factors,x=~effect.sizes,
          color=~p.values,colors=grDevices::colorRamp(c("darkred","gray")),
          error_x=list(array=effect.errors,color=mycolors),split=factors,showlegend=FALSE,marker=list(color=mycolors)) %>%
      layout(xaxis=list(title="Effect Size",zeroline=T,showticklabels=T),yaxis=yform)

Just odd that this first point was so difficult to solve and results in such a big code bracket, because normally plotly supports that pipe logic quite well and you get a very readable code with all the add -functions.

I expected eg, some add_errorbar -function, but apparently you have to add the errorbars in the plot_ly -function and the color-vector for the errors only works if you use the split -function. If someone would like to comment or post an alternative answer with more readable code on this, that would be interesting.

Answer 2

Here is an idea by constructing first a ggplot2 graph and using ggplotly :

create a data frame :

df <- data.frame(factors = factor(factors, levels = factors), #note the order of the levels which determines the order of the y axes
                 effect.sizes = effect.sizes,
                 effect.errors = effect.errors,
                 p.values = p.values)

create the ggplot graph:

library(ggplot2)
library(plotly)

ggplot(df)+
  geom_vline(xintercept = 0, color = "grey50") +
  geom_point(aes(y = factors,
                 x = effect.sizes,
                 color = p.values)) +
  geom_errorbarh(aes(y = factors,
                     xmin = effect.sizes - effect.errors,
                     xmax = effect.sizes + effect.errors,
                     x = effect.sizes,
                     color = p.values)) +
  scale_color_continuous(low = "darkred", high = "gray")+
  theme_bw() +
  xlab("Effect Sizes")+
  ylab("Factors") +
  theme(panel.border = element_blank(),
        plot.margin = margin(1, 1, 1, 1, "cm")) -> p1


ggplotly(p1)

data:

set.seed(1)
factors <- paste0(1:25,":age")
effect.sizes <- rnorm(25,0,1)
effect.errors <- abs(rnorm(25,0,1))
p.values <- runif(25,0,1)