Data Organisation and Introduction to R

Cancer Research Uk Bioinformatics Winter School: Cambridge, 7th-9th December 2016

• Session 1 - Data Organisation (in spreadsheets)

• Session 2 - Introduction to R and Rstudio

• Session 3 - Data manipulation in R

• Session 4 - Data visualisation in R (short intro - more on Day 2)

We will use material from the lessons developed by the Data Carpentry initiative.

This is not a full Data Carpentry course, just a shortened version of it (but we do run these at the University, if you're interested).

• Recognise and apply principles and good practices for organising your data
• Recognise common pitfalls when recording data (and learn to avoid them!)
• Understanding file formats and their suitability for other programs (e.g. R)

The material for this session is available from the Data Carpentry website ( here).

Let's say I recorded some collection dates in the form “Month-Day”.

I didn't bother with the year, because I know what year I'm in right now!

Input these data in a worksheet:

Feb-12
Mar-14
Apr-21

What Day, Month, Year is assumed by your spreadsheet program?

Be careful, different spreadsheet programs assume different things!

Even different versions of the same software might assume different things!

• To avoid problems, store dates as separate variables with the Day, Month, Year

More about dates here.

If possible, store/distribute data in CSV (comma-separated-values) format:

species,year,month,day,weight_kg,height_cm
mouse,2014,3,21,2,10
dog,2013,7,2,20,60
cat,2016,12,7,4.2,25

More about exporting data here.

• Getting to know Rstudio
• Learn basic syntax of R

The materials for this session are available from the Data Carpentry website (Lesson 1 here).

• Read data from a file
• Understand what factors are
• Understand how to manipulate a data.frame object
• Filter and summarise data

The materials for this session are available from the Data Carpentry website (Lessons 2-4 here).

• Learn about 3 types of plots: scatterplots, histograms, boxplots
• Make some customisations to produce clearer plots

We will no longer use the Data Carpentry lessons, because we will use base R graphics (the default in R).

However, if you want to learn about an alternative, please check the Data Carpentry lesson on using the ggplot2 package.

Make sure you have the surveys data in the right format (as detailed in this lesson):

library(dplyr)
surveys_complete <- surveys %>%
  filter(species_id != "",         # remove missing species_id
         !is.na(weight),           # remove missing weight
         !is.na(hindfoot_length),  # remove missing hindfoot_length
         sex != "")                # remove missing sex

## Extract the most common species_id
species_counts <- surveys_complete %>%
                  group_by(species_id) %>%
                  tally %>%
                  filter(n >= 50) %>%
                  select(species_id)

## Only keep the most common species
surveys_complete <- surveys_complete %>%
                 filter(species_id %in% species_counts$species_id)

Let's say we want to see if there is a correlation between the animal's weight and its hindfoot length. We can use the plot() function for this, indicating what values we want to be assigned to the x and y axis.

plot(x = surveys_complete$weight, y = surveys_complete$hindfoot_length)

Challenge: how can we change the axis labels? Check the help for the function with ?plot

plot(surveys_complete$weight, surveys_complete$hindfoot_length,
     xlab = "Weight (g)", ylab = "Hindfoot length (mm)",
     pch = 16, col = "seagreen")

What do the pch and col arguments do?

Many of these graphical parameters can be found in the par function (see its help: ?par).

We would like to know what kind of distribution our data follow.

A typical way to display this information is using a histogram:

hist(surveys_complete$hindfoot_length, breaks = 100, xlab = "Hindfoot length (mm)")

What does this distribution suggest?

The previous histogram suggests that we have several distributions, which is likely because we have weights from different species. We can visualise the distributions of the weight within each species using a boxplot.

The syntax for this is special:

boxplot(surveys_complete$hindfoot_length ~ surveys_complete$species_id)

We will learn more about the special '~' tomorrow, but for now it means that we want to plot the hindfoot length split by species id.

boxplot(surveys_complete$hindfoot_length ~ surveys_complete$species_id, pch = 16, las = 2)

Why are there so many species with no data?

(also, notice the new las argument. Check ?par to know more about it.)

Even though we filtered our data earlier, the levels of our factor variables remained unchanged.

levels(surveys_complete$species_id)

 [1] "AB" "AH" "AS" "BA" "CB" "CM" "CQ" "CS" "CT" "CU" "CV" "DM" "DO" "DS"
[15] "DX" "NL" "OL" "OT" "OX" "PB" "PC" "PE" "PF" "PG" "PH" "PI" "PL" "PM"
[29] "PP" "PU" "PX" "RF" "RM" "RO" "RX" "SA" "SC" "SF" "SH" "SO" "SS" "ST"
[43] "SU" "UL" "UP" "UR" "US" "ZL"

levels(surveys$species_id)

 [1] "AB" "AH" "AS" "BA" "CB" "CM" "CQ" "CS" "CT" "CU" "CV" "DM" "DO" "DS"
[15] "DX" "NL" "OL" "OT" "OX" "PB" "PC" "PE" "PF" "PG" "PH" "PI" "PL" "PM"
[29] "PP" "PU" "PX" "RF" "RM" "RO" "RX" "SA" "SC" "SF" "SH" "SO" "SS" "ST"
[43] "SU" "UL" "UP" "UR" "US" "ZL"

If we want to remove these levels, we can use the function droplevels():

surveys_complete <- droplevels(surveys_complete)
levels(surveys_complete$species_id)

 [1] "DM" "DO" "DS" "NL" "OL" "OT" "PB" "PE" "PF" "PM" "PP" "RF" "RM" "SH"

boxplot(surveys_complete$hindfoot_length ~ surveys_complete$species_id, pch = 16, las = 1)