practical ggplot2

Scatter plots

The mtcars dataset is provided by the ggplot2 library (have a look above at the first lines printed using the head() function). As for every (or near to every) function, most datasets shipped with a library contain also a useful help page (?).

1. Plot the fuel consumption on the y axis and the cars weight on the x axis.

2. The geom_smooth() layer can be used to add a trend line. Try to overlay it to your scatter plot.

3. It would be more useful to draw a regression line on the scatter plot and without the standard error. Using the help (?), adjust the relevant setting in geom_smooth().

4. Ajust some aesthetics in order to:
   • associate the shape of the points to the number of cylinders
   • associate a colour gradient to the quarter mile time
   • If you’re feeling succesful, try to adjust the colour gradient from yellow for fast accelerating cars to blue for slow accelerating cars.
   • have a look at the R package viridis which provides palettes color-blind friendly and suitable for grey printing

5. Find a way to produce both of the following plots:

Categorical data

We will now look at another built-in dataset called ToothGrowth. This dataset contains the teeth length of 60 guinea pigs which received 3 different doses of vitamin C (in mg/day), delivered either by orange juice (OJ) or ascorbic acid (VC).

6. Is this dataset tidy?

7. plot the distributions as boxplots of the teeth lengths by the dose received

8. attribute a different filling colour to each delivery method

When the dataset is tidy, it is easy to draw a plot telling us the story: vitamin C affects the teeth growth and the delivery method is only important for lower concentrations.

Boxplots are nice but misleading. The size of the dataset is not visible and the shapes of distrubutions could be better represented.

9. plot again length of the teeth by the received dose but using geom_dotplot()

10. add a geom_violin() to the previous plot to get a better view of the distribution shape.

Now we are missing summary values like the median which is shown by the boxplots. We should add one.

11. add the median using stat_summary() to the previous plot

next year replaces stat_summary(fun.y = "median", geom = "point", colour = "red", size = 4) by geom_point(stat = "summary", fun.y = "median", colour = "red", size = 4)

12. change the stat_summary() in the previous plot from median to mean_cl_boot and polish the labels.

different summary statistics from the library Hmisc are available. Let’s try the mean_cl_boot that computes the non-parametric bootstrap to obtain 95% confidence intervals (mean_cl_normal assumes normality)

Of note, a ggplot extension named ggbeeswarm proposes a very neat dotplot that fits the distribution.

ToothGrowth %>%
  ggplot(aes(x = factor(dose), y = len)) +
  ggbeeswarm::geom_quasirandom()

Supplementary exercises: genome-wide copy number variants (CNV) detection

Let’s have a look at a real output file for CNV detection. The used tool is called Reference Coverage Profiles: RCP. It was developed by analyzing the depth of coverage in over 6000 high quality (>40×) genomes. In the end, for every kb a state is assigned and similar states are merged eventually.

state means:

• 0, no coverage
• 1, deletion
• 2, expected diploidy
• 3, duplication
• 4, > 3 copies

Reading data

The file is accessible here. It is gzipped but readr will take care of the decompression. Actually, readr can even read the file directly from the website so you don’t need to download it locally.

CNV.seg.gz has 5 columns and the first 10 lines look like:

CNV

13. load the file CNV.seg.gz in R.

Warning

several issues must be fixed:

• comment should be discarded.
• chromosome will be read as integers since first 1000 lines are 1. But, X, Y are at the file’s end.
• first and last column names are unclean. #chrom comtains a hash and length (kb). Would be neater to fix this upfront.