Putting It All Together: Advanced Plotting

Introduction to R For Biologists and Bioinformatics

Dr Stevie Pederson

Black Ochre Data Labs
Telethon Kids Institute

September 19, 2023

Using the
Tidyverse
Together

Using the Tidyverse

We’ve already learned a huge amount:

Importing Data using readr
Making plots with ggplot2
Working with vectors and logical tests
Using dplyr with spreadsheet-like data
Joining functions together with |>

Next we’ll:
- Use everything to make more detailed figures
- Perform a few analyses
- Utilise some features of tidyr

Section Setup

Clear your R Environment
Start a new R script: AdvancedPlotting.R
Load the tidyverse
Define a default ggplot2 theme
Import Transport

library(tidyverse)
theme_set(theme_bw())
cols <- c(
    "gender", "name", "weight", "height", "method"
)
transport <- "data/transport.csv" |>
    read_csv(
        comment = "#",
        col_names = cols, 
        col_types = "-ccnnc"
    ) |>
    mutate(
        gender = case_when(
            gender == "F" ~ "female",
            gender == "M" ~ "male",
            TRUE ~ str_to_lower(gender)
        ),
        gender = as.factor(gender),
        method = factor(
            method, 
            levels = c("car", "bike")
        ),
        BMI = weight / (0.01 * height) ^ 2
    )

Section Outline

Using the pipe to create plots
Customising plots in detail
Reshaping data using tidyr

Visualising Data

Visualising Our Data

What might we like to show?
- Relationship between weight & height?
- Association with transportation method
- Distributions of BMI?

A Simple X-Y Plot

When plotting, we can simply pipe the data into ggplot

transport |>
    ggplot(aes(height, weight)) +
    geom_point()

Adding Lines of Best Fit

Sometimes a regression line can be informative
- geom_smooth() guesses the best line
- method = 'loess' isn’t that great here

transport |>
    ggplot(aes(height, weight)) +
    geom_point() +
    geom_smooth()

Adding Regresion Lines

We can choose a linear regression line: method = "lm"
- We fit linear regression in R using the function lm()
- Hide the standard error of the regression line: se = FALSE

transport |>
    ggplot(aes(height, weight)) +
    geom_point() +
    geom_smooth(method = "lm", se =  FALSE)

Customising Parameters

We can change the colour of the line: colour = "black"
- Anything set inside aes() should be a column in your data
- Anything set outside of aes() should be a fixed-value
- Can also set linetype, linewidth, alpha etc

transport |>
    ggplot(aes(height, weight)) +
    geom_point() +
    geom_smooth(
        method = "lm", se =  FALSE,
        colour = "black"
    )

Changing Shapes

Similarly for the points: colour = "grey30"
- The range of shapes is visible using ?pch
- For shapes 21-25 colour is outline, fill is the internal colour

transport |>
    ggplot(aes(height, weight)) +
    geom_point(colour = "grey30", shape = 1) +
    geom_smooth(
        method = "lm", se =  FALSE,
        colour = "black"
    )

Changing Shapes

We can also set any character to be the point shape
- Additional parameters include size, alpha

transport |>
    ggplot(aes(height, weight)) +
    geom_point(
        colour = "grey30", shape = "#", size = 4
    ) +
    geom_smooth(
        method = "lm", se =  FALSE,
        colour = "black"
    )

Parameters Inside or Outside `aes()`

Parameters set outside aes() will over-ride anything inside aes()
- We have globally set colour to depend on gender
- This is overridden by both geom_point() and geom_smooth()

transport |>
    ggplot(
        aes(height, weight, colour = gender)
    ) +
    geom_point(
        colour = "grey30", shape = "#", size = 4
    ) +
    geom_smooth(
        method = "lm", se =  FALSE,
        colour = "black"
    )

Parameters Inside or Outside `aes()`

Now remove the colour from geom_point()
- Inherits from the aes() within ggplot()

transport |>
    ggplot(
        aes(height, weight, colour = gender)
    ) +
    geom_point(shape = "#", size = 4) +
    geom_smooth(
        method = "lm", se =  FALSE,
        colour = "black"
    )

Custom Scales

Providing specific colours can take a vector
- Can be named for greater control

transport |>
    ggplot(
        aes(height, weight, colour = gender)
    ) +
    geom_point(shape = "#", size = 4) +
    geom_smooth(
        method = "lm", se =  FALSE,
        colour = "black"
    ) +
    scale_colour_manual(
        values = c(
            female = "navyblue", male = "red3"
        )
    )

Custom Scales

Likewise for shapes
- Values are applied in the same order as the legend

transport |>
    ggplot(
        aes(height, weight, colour = gender)
    ) +
    geom_point(
        aes(shape = gender), size = 4
    ) +
    geom_smooth(
        method = "lm", se =  FALSE,
        colour = "black"
    ) +
    scale_colour_manual(
        values = c(
            female = "navyblue", male = "red3"
        )
    ) +
    scale_shape_manual(
        values = c("F", "M")
    )

Adding Statistics

Sometimes we might wish to add summary statistics to plots
We can create on the fly inside a geom_

transport |>
    ggplot(aes(height, weight)) +
    geom_point(
        aes(colour = gender, shape = gender),
        size = 3
    ) +
    geom_smooth(
        method = "lm", se =  FALSE,
        colour = "black"
    ) +
    geom_label(
        aes(label = label),
        data = . %>% ## Only the magrittr works here
            summarise(
                cor = cor(weight, height),
                height = mean(height),
                ## Specify a position manually
                # height = 165, 
                weight = min(weight),
            )  %>% 
            mutate(
                label = paste("rho ==", round(cor, 2))
            ),
        parse = TRUE
    ) +
    scale_colour_manual(
        values = c(
            female = "navyblue", male = "red3"
        )
    ) +
    scale_shape_manual(
        values = c("F", "M")
    )

Using Additional Packages

There are multiple options for adding other labels

Add this to the top of your script
\(\implies\) Make sure the package is loaded

library(ggpmisc)

Using Additional Packages

stat_poly_eq() can add \(R^2\), adjusted \(R^2\) or regression equations

transport |>
    ggplot(aes(height, weight)) +
    geom_point(
        aes(colour = gender, shape = gender), 
        size = 3
    ) +
    geom_smooth(
        method = "lm", se =  FALSE,
        colour = "black"
    ) +
    stat_poly_eq(use_label("eq")) +
    scale_colour_manual(
        values = c(
            female = "navyblue", male = "red3"
        )
    ) +
    scale_shape_manual(
        values = c("F", "M")
    )

Multiple Regression Equations

Combining with facets can provide multiple equations

transport |>
    ggplot(aes(height, weight)) +
    geom_point(
        aes(colour = gender, shape = gender), 
        size = 3
    ) +
    geom_smooth(
        method = "lm", se =  FALSE,
        colour = "black"
    ) +
    stat_poly_eq(use_label("eq")) +
    facet_wrap(~method) +
    scale_colour_manual(
        values = c(
            female = "navyblue", male = "red3"
        )
    ) +
    scale_shape_manual(
        values = c("F", "M")
    )

Modifying Data
Prior To Plotting

Summary Plots

So far we’ve just plotted the complete dataset
- We did use %>% inside a geom_* to find a correlation…

We can also use our tidyverse tools to create summaries to plot
- E.g. Barplots of the mean with error bars

Summary Plots

To create a barplot of the mean BMI across all groups
- Need the mean BMI for each group
- Also the standard deviation

transport |>
    summarise(
        mn_bmi = mean(BMI),
        sd_bmi = sd(BMI),
        .by = c(method, gender)
    )

Creating a Barplot

Now we can create a barplot using geom_col()

transport |>
    summarise(
        mn_bmi = mean(BMI), sd_bmi = sd(BMI),
        .by = c(method, gender)
    ) |>
    ggplot(aes(method, mn_bmi, fill = gender)) +
    geom_col() +
    facet_wrap(~gender) +
    labs(y = "Mean BMI") +
    scale_fill_brewer(palette = "Set2")

Adding Error Bars

We add error bars using geom_errorbar()

transport |>
    summarise(
        mn_bmi = mean(BMI), sd_bmi = sd(BMI),
        .by = c(method, gender)
    ) |>
    ggplot(aes(method, mn_bmi, fill = gender)) +
    geom_col() +
    geom_errorbar(
        aes(
            ymin = mn_bmi - sd_bmi,
            ymax = mn_bmi + sd_bmi
        ),
        width = 0.2
    ) +
    facet_wrap(~gender) +
    labs(y = "Mean BMI") +
    scale_fill_brewer(palette = "Set2")

Adding Error Bars

If we choose not to facet it’s much trickier
- Let’s hide the error bars to see why

transport |>
    summarise(
        mn_bmi = mean(BMI), sd_bmi = sd(BMI),
        .by = c(method, gender)
    ) |>
    ggplot(aes(method, mn_bmi, fill = gender)) +
    geom_col(position = "dodge") +
    labs(y = "Mean BMI") +
    scale_fill_brewer(palette = "Set2")

Adding Error Bars

Adding position = "dodge" separates the bars horizontally
What is the x co-ordinate for each bar now?
What is the width of each bar?

Plotting with Factors

Let’s try coerce factors to integers

transport |>
    summarise(
        mn_bmi = mean(BMI), sd_bmi = sd(BMI),
        .by = c(method, gender)
    ) |>
    mutate(
        method_int = as.integer(method),
        gender_int = as.integer(gender)
    )

# A tibble: 4 × 6
  method gender mn_bmi sd_bmi method_int gender_int
  <fct>  <fct>   <dbl>  <dbl>      <int>      <int>
1 car    female   23.0  1.14           1          1
2 bike   female   23.3  0.961          2          1
3 bike   male     25.7  1.47           2          2
4 car    male     25.8  0.799          1          2

Plotting with Factors

Each factor level is assigned an integer starting at 1
- These were how ggplot originally placed the x co-ordinates

Can we figure out our x-co-ordinates yet?

I guesstimated that each bar had a width of 0.45
- The centre of each bar is method_int \(\pm\) 0.225

I also wish it was easier
- Maybe I just haven’t found the right package/function yet

Plotting with Factors

transport |>
    summarise(
        mn_bmi = mean(BMI), sd_bmi = sd(BMI),
        .by = c(method, gender)
    ) |>
    mutate(
        method_int = as.integer(method),
        gender_int = as.integer(gender),
        x = case_when(
            gender == "female" ~ method_int - 0.225,
            gender == "male" ~ method_int + 0.225,
        )
    )

# A tibble: 4 × 7
  method gender mn_bmi sd_bmi method_int gender_int     x
  <fct>  <fct>   <dbl>  <dbl>      <int>      <int> <dbl>
1 car    female   23.0  1.14           1          1 0.775
2 bike   female   23.3  0.961          2          1 1.78 
3 bike   male     25.7  1.47           2          2 2.22 
4 car    male     25.8  0.799          1          2 1.23

Plotting with Factors

transport |>
    summarise(
        mn_bmi = mean(BMI), sd_bmi = sd(BMI),
        .by = c(method, gender)
    ) |>
    mutate(
        method_int = as.integer(method),
        x = case_when(
            gender == "female" ~ method_int - 0.225,
            gender == "male" ~ method_int + 0.225,
        )
    ) |>
    ggplot(aes(method, mn_bmi, fill = gender)) +
    geom_col(position = "dodge") +
    geom_errorbar(
        aes(
            x = x,
            ymin = mn_bmi - sd_bmi,
            ymax = mn_bmi + sd_bmi
        ),
        width = 0.1
    ) +
    labs(y = "Mean BMI") +
    scale_fill_brewer(palette = "Set2")

Modifying Axes

That space at the bottom of the y-axis bother me
- By default axes expand the data by ~5% of the range
We can control the expansion
- Using multiplicative scaling (continuous data)
- Using Additive Scaling (discrete data)
The relevant function is called expansion()
- Is passed to the expand argument inside scale_x/y_*()

Modifying Axes

We also label them using the name argument

transport |>
    summarise(
        mn_bmi = mean(BMI), sd_bmi = sd(BMI),
        .by = c(method, gender)
    ) |>
    mutate(
        method_int = as.integer(method),
        x = case_when(
            gender == "female" ~ method_int - 0.225,
            gender == "male" ~ method_int + 0.225,
        )
    ) |>
    ggplot(aes(method, mn_bmi, fill = gender)) +
    geom_col(position = "dodge") +
    geom_errorbar(
        aes(
            x = x,
            ymin = mn_bmi - sd_bmi,
            ymax = mn_bmi + sd_bmi
        ), width = 0.1
    ) +
    scale_fill_brewer(
        palette = "Set2", name = "Gender"
    ) +
    scale_y_continuous(
        expand = expansion(c(0, 0.05)), 
        name = "Mean BMI"
    )

More Complex Summaries

So far we only visualised the mean BMI values across the dataset
Let’s try and repeat for both weight and height on the same figure
- May not be brilliant in reality

transport |>
    summarise(
        across(
            ends_with("ght"),
            .fns = c(mn = mean, sd = sd)
        ),
        .by = c(method, gender)
    )

# A tibble: 4 × 6
  method gender weight_mn weight_sd height_mn height_sd
  <fct>  <fct>      <dbl>     <dbl>     <dbl>     <dbl>
1 car    female      61.4      5.08      163.      3.20
2 bike   female      65.3      5.66      167.      3.95
3 bike   male        80.8     10.1       177.      5.77
4 car    male        81.1      6.44      177.      4.45

Can we get the means into one column and sd into another?

Reshaping Data

pivot_longer() enables shifting from wide to long form
Choose the columns to run down the page
- Original column names go into one column
- All values go into another

transport |>
    summarise(
        across(
            ends_with("ght"),
            .fns = c(mn = mean, sd = sd)
        ),
        .by = c(method, gender)
    ) |>
    pivot_longer(cols = contains("_"))

# A tibble: 16 × 4
   method gender name       value
   <fct>  <fct>  <chr>      <dbl>
 1 car    female weight_mn  61.4 
 2 car    female weight_sd   5.08
 3 car    female height_mn 163.  
 4 car    female height_sd   3.20
 5 bike   female weight_mn  65.3 
 6 bike   female weight_sd   5.66
 7 bike   female height_mn 167.  
 8 bike   female height_sd   3.95
 9 bike   male   weight_mn  80.8 
10 bike   male   weight_sd  10.1 
11 bike   male   height_mn 177.  
12 bike   male   height_sd   5.77
13 car    male   weight_mn  81.1 
14 car    male   weight_sd   6.44
15 car    male   height_mn 177.  
16 car    male   height_sd   4.45

Reshaping Data

We can now separate() the values in name
- into = c("stat", "type") defines the new column names

transport |>
    summarise(
        across(
            ends_with("ght"),
            .fns = c(mn = mean, sd = sd)
        ),
        .by = c(method, gender)
    ) |>
    pivot_longer(cols = contains("_")) |>
    separate(
        name, into = c("stat", "type"), 
        sep = "_"
    )

# A tibble: 16 × 5
   method gender stat   type   value
   <fct>  <fct>  <chr>  <chr>  <dbl>
 1 car    female weight mn     61.4 
 2 car    female weight sd      5.08
 3 car    female height mn    163.  
 4 car    female height sd      3.20
 5 bike   female weight mn     65.3 
 6 bike   female weight sd      5.66
 7 bike   female height mn    167.  
 8 bike   female height sd      3.95
 9 bike   male   weight mn     80.8 
10 bike   male   weight sd     10.1 
11 bike   male   height mn    177.  
12 bike   male   height sd      5.77
13 car    male   weight mn     81.1 
14 car    male   weight sd      6.44
15 car    male   height mn    177.  
16 car    male   height sd      4.45

Reshaping Data

The opposite of pivot_longer() is pivot_wider()

transport |>
    summarise(
        across(
            ends_with("ght"),
            .fns = c(mn = mean, sd = sd)
        ),
        .by = c(method, gender)
    ) |>
    pivot_longer(cols = contains("_")) |>
    separate(
        name, into = c("stat", "type"), 
        sep = "_"
    ) |>
    pivot_wider(
        names_from = "type",
        values_from = "value"
    )

# A tibble: 8 × 5
  method gender stat      mn    sd
  <fct>  <fct>  <chr>  <dbl> <dbl>
1 car    female weight  61.4  5.08
2 car    female height 163.   3.20
3 bike   female weight  65.3  5.66
4 bike   female height 167.   3.95
5 bike   male   weight  80.8 10.1 
6 bike   male   height 177.   5.77
7 car    male   weight  81.1  6.44
8 car    male   height 177.   4.45

Now we have something we can use for barplots with errorbars

Reshaping Data

For the sake of brevity going forward, save that as summ_df

summ_df <- transport |>
    summarise(
        across(
            ends_with("ght"),
            .fns = c(mn = mean, sd = sd)
        ),
        .by = c(method, gender)
    ) |>
    pivot_longer(cols = contains("_")) |>
    separate(
        name, into = c("stat", "type"), 
        sep = "_"
    ) |>
    pivot_wider(
        names_from = "type",
        values_from = "value"
    )

Plotting Multiple Summaries

Hide the y-axis name using name = c()

summ_df |>
    mutate(
        gender_int = as.numeric(gender),
        x_bar = case_when(
            method == "car" ~ gender_int - 0.225,
            method == "bike" ~ gender_int + 0.225
        )
    ) |>
    ggplot(aes(gender, mn, fill = method)) +
    geom_col(position = "dodge") +
    geom_errorbar(
        aes(
            x = x_bar,
            ymin = mn - sd, ymax = mn + sd
        ),
        width = 0.2
    )  +
    facet_wrap(~stat, scales = "free_y") +
    scale_fill_brewer(
        palette = "Set2", name = "Method"
    ) +
    scale_y_continuous(
        expand = expansion(c(0, 0.05)), 
        name = c()
    )

Plotting Multiple Summaries

Add the units to the facet names

summ_df |>
    mutate(
        gender_int = as.numeric(gender),
        x_bar = case_when(
            method == "car" ~ gender_int - 0.225,
            method == "bike" ~ gender_int + 0.225
        ),
        stat = stat |>
            str_replace("height", "Height (cm)") |>
            str_replace("weight", "Weight (kg)")
    ) |>
    ggplot(aes(gender, mn, fill = method)) +
    geom_col(position = "dodge") +
    geom_errorbar(
        aes(
            x = x_bar,
            ymin = mn - sd, ymax = mn + sd
        ),
        width = 0.2
    )  +
    facet_wrap(~stat, scales = "free_y") +
    scale_fill_brewer(
        palette = "Set2", name = "Method"
    ) +
    scale_y_continuous(
        expand = expansion(c(0, 0.05)), name = c()
    )

Themes

Using Themes

The overall plot appearance can be set using theme()
A bit overwhelming, but very customisable
Let’s save our existing plot a p
- Can be drawn again at anytime by entering p
- Can still be modified on on-the-fly

p <- summ_df |>
    mutate(
        gender_int = as.numeric(gender),
        x_bar = case_when(
            method == "car" ~ gender_int - 0.225,
            method == "bike" ~ gender_int + 0.225
        ),
        stat = stat |>
            str_replace("height", "Height (cm)") |>
            str_replace("weight", "Weight (kg)")
    ) |>
    ggplot(aes(gender, mn, fill = method)) +
    geom_col(position = "dodge") +
    geom_errorbar(
        aes(
            x = x_bar,
            ymin = mn - sd, ymax = mn + sd
        ),
        width = 0.2
    )  +
    facet_wrap(~stat, scales = "free_y") +
    xlab("Gender") +
    scale_fill_brewer(
        palette = "Set2", name = "Method"
    ) +
    scale_y_continuous(
        expand = expansion(c(0, 0.05)), name = c()
    )

Using Themes

If we try to add a title, it’s aligned left

p + ggtitle("Weight and Height Across Participants")

Using Themes

Most theme() elements have their own modification functions
- text: element_text()
- lines: element_line()
- rectangles: element_rect()

All can be removed using element_blank

The plot title needs element_text()

Using Themes

hjust controls the horizontal adjustment
- hjust = 0.5 is centre-aligned

p + 
    ggtitle("Weight and Height Across Participants") +
    theme(plot.title = element_text(hjust = 0.5))

Using Themes

We can resize all primary text

p + 
    ggtitle("Weight and Height Across Participants") +
    theme(
        text = element_text(size = 14),
        plot.title = element_text(hjust = 0.5)
    )

Using Themes

Control legend position
- Doesn’t need an element_*() function

p + 
    ggtitle("Weight and Height Across Participants") +
    theme(
        text = element_text(size = 14),
        plot.title = element_text(hjust = 0.5),
        legend.position = "bottom"
    )

Using Themes

Hide the background grid & rotates x-axis text

p + 
    ggtitle("Weight and Height Across Participants") +
    theme(
        text = element_text(size = 14),
        plot.title = element_text(hjust = 0.5),
        legend.position = "bottom",
        panel.grid = element_blank(),
        axis.text.x = element_text(
            angle = 90, vjust = 0.5, hjust = 1
        )
    )

Exporting Plots

Making plots in R is nice
\(\implies\) How do we get them into our paper??!!!

ggsave() will save the last plot
- The file extension will determine the format
- Can be png, jpg, pdf, svg, tiff etc
- width & height default to inches but can be changed

Getting font sizes right can be infuriating
- Always add the save after you create the plot
- Open immediately and check the font sizes

Closing Comments

Can now (hopefully) make the figures for our next paper
ggplot2 is very powerful \(\implies\) takes a long time to master
Getting data structured correctly is an important part

Note that once we loaded data \(\implies\) never modified

We saved four objects
- cols, transport, summ_df, p
- The last two were only to fit the code on slides

This keeps a clean workspace
- No need for transport, transport1, transport1_mod etc
- Very beneficial for reproducibility

Putting It All Together: Advanced Plotting

Using theTidyverseTogether

Using the Tidyverse

Section Setup

Section Outline

Visualising Data

Visualising Our Data

A Simple X-Y Plot

Adding Lines of Best Fit

Adding Regresion Lines

Customising Parameters

Changing Shapes

Changing Shapes

Parameters Inside or Outside aes()

Parameters Inside or Outside aes()

Custom Scales

Custom Scales

Adding Statistics

Using Additional Packages

Using Additional Packages

Multiple Regression Equations

Modifying DataPrior To Plotting

Summary Plots

Summary Plots

Creating a Barplot

Adding Error Bars

Adding Error Bars

Adding Error Bars

Plotting with Factors

Plotting with Factors

Plotting with Factors

Plotting with Factors

Modifying Axes

Modifying Axes

More Complex Summaries

Reshaping Data

Reshaping Data

Reshaping Data

Reshaping Data

Plotting Multiple Summaries

Plotting Multiple Summaries

Themes

Using Themes

Using Themes

Using Themes

Using Themes

Using Themes

Using Themes

Using Themes

Exporting Plots

Closing Comments

Using the
Tidyverse
Together

Parameters Inside or Outside `aes()`

Parameters Inside or Outside `aes()`

Modifying Data
Prior To Plotting