Data Carpentry for Biologists

Graphing using ggplot

Notes

Introduction to the UHURU dataset

Data

  • For the next set of lessons we’ll be working with data on acacia size from an experiment in Kenya
  • The experiment is designed to understand the influence of herbivores on vegetation by excluding different sized herbivores
  • There are 3 different treatments:
    • The top-left image shows Megaherbivore exclosures, which use wires 2m high to keep elephants
    • The top-right image shows Mesoherbivore exclosures, which use fenses starting 1/3m off the ground to exclude things like impalla
    • The bottom-left image shows full exclosures, which use fenses all the way to the ground to keep out all mammalian herbivores
    • And the bottom-right image shows control plots

Pictures of 4 experimental treatments. (A) Megaherbivore fences consist of two parallel wires starting 2-m above ground level. (B) Mesoherbivore fences consist of 11 parallel wires starting ~0.3 m above ground level and continuing to 2.4-m above ground level. (C) Total-exclusion fences consist of 14 wires up to 2.4-m above ground level, with a 1-m high chain-link barrier at ground level. (D) Open control plots are unfenced, with boundaries demarcated by short wooden posts at 10-m intervals.

  • So far we’ve been working with datasets that are separated by commas
  • Click on surveys.csv to show csv format
  • But if we look at this new dataset it looks different
    • Click on Acacia Dataset to Open in Text Editor
  • This data is tab separated, so we’ll need to treat it differently when we import it
  • To do this we manually set the character separating each column using the optional sep argument
  • So we’ll call our data frame acacia and assign it the output from read.csv
  • We still give it the name of the file in quotes as the first argument
  • Then we add a comma and sep = "\t"
  • \t is who we indicate a Tab character in programming
acacia <- read.csv("ACACIA_DREPANOLOBIUM_SURVEY.txt", sep="\t")
  • We can also see that it includes information on whether or not the plant is dead in the HEIGHT column
  • Is that good data structure?
  • If you said “No”, you’re right, information on if the tree is dead should be stored in a separate column
  • For now, we’ll just treat the “dead” entries as null values
  • We can do this by using another optional argument na.strings
  • So let’s modify our read.csv statement by adding na.strings = c("dead").
  • This will replace the string "dead" with NA
  • It gets passed as a vector because this allows multiple different values to be set as nulls
acacia <- read.csv("ACACIA_DREPANOLOBIUM_SURVEY.txt", sep="\t", na.strings = c("dead"))
  • If we open the resulting table we can see that it includes information on:
    • the time and location of the sampling
    • the experimental treatment
    • the size of each Acacia including a height, the canopy diameter measured in the direction (or axis) or the largest diameter and the diameter of the axis perpendicular to that, and the circumference of the shrub
    • information on the number of flowers, buds, and fruits
    • And finally information on the species of ant associated with the shrub because there is a very interesting ant-acacia mutualism where the Acacia special structures that serve as houses for the ants and the ants swarm herbivores that try to eat the acacia

ggplot

  • Very popular plotting package
  • Good plots quickly
  • Declarative - describe what you want not how to build it
  • Contrasts w/Imperative - how to build it step by step
  • Install ggplot2 using install.packages

Basics

  • We load the ggplot2 package just like we loaded dplyr
library(ggplot2)
  • We’ll also load the UHURU like we discussed in the video on the dataset
acacia <- read.csv("ACACIA_DREPANOLOBIUM_SURVEY.txt", sep="\t", na.strings = c("dead"))
  • To build a plot using ggplot we start with the ggplot() function
ggplot()
  • ggplot() creates a base ggplot object that we can then add things to

  • Like a blank canvas

  • We can also add optional arguments for information to be shared across different components of the plot
  • The two main arguments we typically use here are
  • data - which is the name of the data frame we are working with, so acacia
  • mapping - which describes which columns of the data are used for different aspects of the plot
  • We create a mapping by using the aes function, which stands for “aesthetic”, and then linking columns to pieces of the plot
  • We’ll start with telling ggplot what value should be on the x and y axes
  • Let’s plot the relationship betwen the circumference of an acacia and its height
ggplot(data = acacia, mapping = aes(x = CIRC, y = HEIGHT))
  • This still doesn’t create a figure, it’s just a blank canvas and some information on default values for data and mapping columns to pieces of the plot
  • We can add data to the plot using layers
  • We do this by adding a + after the the ggplot function and then adding something called a geom, which stands for geometry
  • To make a scatter plot we use geom_point
ggplot(data = acacia, mapping = aes(x = CIRC, y = HEIGHT)) +
  geom_point()

  • It is standard to hit Enter after the plus so that each layer shows up on its own line

  • To change things about the layer we can pass additional arguments to the geom
  • We can do things like change
    • the size of the points, we’ll set it to 3
    • the color of the points, we’ll set it to "blue"
    • the transparency of the points, which is called alpha, we’ll set it to 0.5
ggplot(data = acacia, mapping = aes(x = CIRC, y = HEIGHT)) +
  geom_point(size = 3, color = "blue", alpha = 0.5)
  • To add labels (like documentation for your graphs!) we use the labs function
ggplot(data = acacia, mapping = aes(x = CIRC, y = HEIGHT)) +
  geom_point(size = 3, color = "blue", alpha = 0.5) +
  labs(x = "Circumference [cm]", y = "Height [m]",
       title = "Acacia Survey at UHURU")

Rescaling axes

ggplot(data = acacia, mapping = aes(x = CIRC, y = HEIGHT)) +
  geom_point(size = 3, color = "blue", alpha = 0.5) +
  scale_y_log10() +
  scale_x_log10()
  • Not changing the data itself, just the presentation of it

Do Tasks 1-2 in Acacia and ants.

Grouping

  • Group on a single graph
  • Look at influence of experimental treatment
ggplot(acacia, aes(x = CIRC, y = HEIGHT, color = TREATMENT)) +
  geom_point(size = 3, alpha = 0.5)
  • Facet specification
ggplot(acacia, aes(x = CIRC, y = HEIGHT)) +
  geom_point(size = 3, alpha = 0.5) +
  facet_wrap(~TREATMENT)
  • Where are all the acacia in the open plots? (eaten?)

Do Tasks 3-4 in Acacia and ants.

Statistical transformations

  • We’ve seen that ggplot makes graphs by combining information on
    • Data
    • Mapping of parts of that data to aspects of the plot
    • A geometric object to represent the data
ggplot(acacia, aes(x = CIRC, y = HEIGHT)) +
  geom_point()

  • Many kinds of geometric object (type geom_ and show completions)

  • Each geom includes a statistical transformations
  • So far we’ve only seen
    • identity: the raw form of the data or no transformation
  • Transformations exist to make things like histograms, bar plots, etc.

  • Occur as defaults in associated geoms

  • To look at the number of acacia in each treatment use a bar plot 
ggplot(acacia, aes(x = TREATMENT)) +
  geom_bar()
  • Uses the transformation stat_count()
    • Counts the number of rows for each treatment
  • To look at the distribution of circumferences in the dataset use a histogram 
ggplot(acacia, aes(x = CIRC)) +
  geom_histogram(fill = "red")
  • Uses stat_bins() for data transformation
    • Splits circumferences into bins and counts rows in each bin
  • Set number of bins or binwidth
ggplot(acacia, aes(x = CIRC)) +
  geom_histogram(fill = "red", bins = 15)

ggplot(acacia, aes(x = CIRC)) +
  geom_histogram(fill = "red", binwidth = 5)
  • These can be combined with all of the other ggplot2 features we’ve learned

Do Tasks 1-2 in Acacia and ants histograms.

Position

  • geom’s also come with a default position
  • In many cases the position is "identity", which just means the object is plotted in the position determined by the data, but not always
  • Let’s remake our histogram, but color the bars by the treatment
ggplot(acacia, aes(x = CIRC, color = TREATMENT)) +
  geom_histogram(binwidth = 5)
  • You can see that the total height of the bars stayed the same
  • That’s because ggplot has just colored the pieces of each bar that correspond to each treatment
  • It does this because the default position for geom_histogram is "stacked", which stacks the bars on top of one another and therefore makes a stacked histogram.
  • If we want separate overlapping histograms then we need to change the position
ggplot(acacia, aes(x = CIRC, color = TREATMENT)) +
  geom_histogram(binwidth = 5, position = "identity")
  • And then add some transparency so we can see all of the histograms
ggplot(acacia, aes(x = CIRC, color = TREATMENT)) +
  geom_histogram(binwidth = 5, position = "identity", alpha = 0.5)

Layers

  • So far we’ve only plotted one layer or geom at a time, but we can combine multiple layers in a single plot
    • ggplot() sets defaults for all layers
    • Can combine multiple layers using +
    • The first geom is plotted first and then additional geoms are layered on top
  • Combine different kinds of layers
  • Add a linear model
ggplot(acacia, aes(x = CIRC, y = HEIGHT)) +
  geom_point() +
  geom_smooth(method = "lm")
  • Both the geom_point layer and the geom_smooth layer use the defaults form ggplot
  • Both use acacia for data and x = CIRC, y = HEIGHT for the aesthetic

  • geom_smooth uses the statistical transformation stat_smooth() to produce a smoothed representation of the data

  • Do this by treatment
ggplot(acacia, aes(x = CIRC, y = HEIGHT, color = TREATMENT)) +
  geom_point() +
  geom_smooth(method = "lm")
  • Because the color aesthetic is the default it is inherited by geom_smooth
  • One set of points and one model for each treatment

Changing values across layers

  • We can also plot data from different columns or even data frames on the same graph
  • To do this we need to better understand how layers and defaults work
  • So far we’ve put all of the information on data and aesthetic mapping into ggplot()
ggplot(data = acacia, mapping = aes(x = CIRC, y = HEIGHT)) +
  geom_point() +
  geom_smooth(method = "lm")
  • This sets the default data frame and aesthetic, which is then used by geom_point() and geom_smooth()
  • Alternatively instead of setting the default we could just give these values directly to geom_point() and `geom_smo
ggplot() +
  geom_point(data = acacia,
             mapping = aes(x = CIRC, y = HEIGHT,
                           color = TREATMENT)) +
  geom_smooth(method = "lm")

  • We can see that this information is no longer shared with other geoms since it is no longer the default, so we’ve asked for a smooth of nothing and so no smoother is shown

  • Can use this combine different aesthetics
  • Make a single model across all treatments while still coloring points
ggplot() +
  geom_point(data = acacia,
             mapping = aes(x = CIRC, y = HEIGHT,
                           color = TREATMENT)) +
  geom_smooth(data = acacia,
              mapping = aes(x = CIRC, y = HEIGHT))
  • color is only set in the aesthethic for the point layer

  • So the smooth layer is made across all x and y values

  • Check if this makes sense to everyone

  • This same sort of change can be used to plot different columns on the same plot by changing the values of x or y
  • If we wanted to plot two different columns as Y we could change the value of y in the aesthetic

  • If we wanted to use data from two different data frames we could change the value of data

  • We can also keep all of the values that are shared as defaults if we want to
ggplot(data = acacia, mapping = aes(x = CIRC, y = HEIGHT)) +
  geom_point(mapping = aes(color = TREATMENT)) +
  geom_smooth(method = "lm")

Do Task 3 in Acacia and ants histograms.

Understanding defaults (optional if students struggling after exercise)

  • Let’s move some things around to help understand where to put different information data and mapping
ggplot(data = acacia, mapping = aes(x = CIRC, y = AXIS1)) +
  geom_point() +
  geom_smooth(method = "lm")
  • What happens if we move data and mapping to one layer
ggplot() +
  geom_point(data = acacia, mapping = aes(x = CIRC, y = AXIS1)) +
  geom_smooth(method = "lm")
  • What about the other layer
ggplot() +
  geom_point() +
  geom_smooth(data = acacia, mapping = aes(x = CIRC, y = AXIS1), method = "lm")
  • What if we move one to each layer (error)
ggplot() +
  geom_point(data = acacia) +
  geom_smooth(mapping = aes(x = CIRC, y = AXIS1), method = "lm")
  • So, each geom needs access to data and mapping
  • If they are not in the geom then the geom uses the defaults ggplot()
  • If we want different data or mapping for different layers then provide different values to different geoms
ggplot(data = acacia, mapping = aes(x = CIRC)) +
  geom_point(mapping = aes(y = AXIS1), color = "black") +
  geom_point(mapping = aes(y = AXIS2), color = "grey")

Grammar of graphics

  • Uniquely describe any plot based on a defined set of information
  • Leland Wilkinson
  • Geometric object(s)
    • Data
    • Mapping
    • Statistical transformation
    • Position (allows you to shift objects, e.g., spread out overlapping data points)
  • Facets
  • Coordinates (coordinate systems other than cartesian, also allows zooming)

Saving plots as new files

ggsave("acacia_by_treatment.jpg")
  • The type of the file is determined by the file extension
ggsave("acacia_by_treatment.pdf")
  • Lots of optional arguments including size
ggsave("acacia_by_treatment.pdf", height = 5, width = 5)

Assign the rest of the exercises.