2.1 Pipes

Rewrite the following code chunks as pipes (magrittr):

my_cars <- mtcars[, c(1:4, 10)]
my_cars <- my_cars[my_cars$disp > mean(my_cars$disp), ]
my_cars <- colMeans(my_cars)

mtcars %>%
  select(c(1:4, 10)) %>%
  filter(disp > mean(disp)) %>%
  colMeans() -> my_cars

# It is better to have the result assigned on the left hand side: result <- expression. In this case the expression is the **whole** pipe.
# Our 'expression -> result' is correct but can easily be missed when reading the code.

summary(cars)
colSums(cars)

cars %T>% {print(summary(.))} %>% colSums()

cor(mtcars$gear, mtcars$mpg)

mtcars %$% cor(gear, mpg)

cor(mtcars)

mtcars %>% cor()

dim_summary <- function(nrows, ncols) {
  print(
    paste0('Matrix M has: ', nrows, ' rows and ', ncols, ' columns.')
  )
}

distr1 <- rnorm(16)
M <- matrix(distr1, ncol = 4)
plot(M)
M <- M + sample(M)
dim_summary(nrows = nrow(M), ncols = ncol(M))
distr2 <- rnorm(16)

N <- matrix(distr2, ncol = 4)
colnames(N, letters[1:4])
summary(N)

P <- M %x% t(N)
heatmap(P)
names(P) <- letters[1:dim(P)[2]]
cor(P[ ,'a'], P[ ,'i'])

dim_summary <- function(nrows, ncols) {
  print(paste0('Matrix M has: ', nrows, ' rows and ', ncols, ' columns.'))
}

M <- rnorm(16) %>%
  matrix(ncol = 4) %T>%
  plot() %>%
  `+`(., sample(.)) %T>%
  {dim_summary(nrow(.), ncol(.))}

N <- rnorm(16) %>%
  matrix(ncol = 4) %>%
  `colnames<-`(letters[1:4]) %>%
  t()

P <- M %>%
  `%x%`(., N) %T>%
  heatmap() %>%
  `colnames<-`(letters[1:dim(.)[2]]) %>%
  as_data_frame() %$%
  cor(a, i)

2.2 Tibbles

# 1
vehicles <- mtcars %>% as_tibble()

# 2
vehicles[['cyl']]
vehicles[[2]]
vehicles$cyl

# 3
# Here we used pipe, you do not have to...
vehicles %T>%
  {print(.[['cyl']])} %T>%
  {print(.[[2]])} %>%
  .$cyl

# 4
vehicles

# 5
vehicles %>% head(n = 30)

# 6
options(tibble.print_min = 15, tibble.print_max = 30)

# 7
# In theory there should be no difference. dplyr imports tibble from the tibble package
# and dplyr.width, dplyr.print_min and dplyr.print_min are passed down to the tibble.
# But test both behaviours. First with only the tibble package loaded, later with dplyr # loaded.

# 8
automobiles <- as.data.frame(vehicles)

tab <- tribble(
  ~id, ~event, ~date,
  1, 'success', '24-04-2017',
  2, 'failed', '25-04-2017',
  3, 'failed', '25-04-2017',
  4, 'success', '27-04-2017'
)

tst <- replicate(30, sample(100), simplify = TRUE)
names(tst) = paste0(rep('col', times = dim(tst)[2]), 1:dim(tst)[2])
microbenchmark::microbenchmark(
  as.data.frame(tst),
  as_data_frame(tst),
  as_tibble(tst)
)

tibble(x = sample(1:10, size = 10, replace = T), y = log10(x))

2.3 Parsing

Parse the following vectors:

• vec1 <- c(1, 7.2, 3.84, -5.23) – parse it as double.
• now parse the same vector c(1, 7.2, 3.84, -5.23) as integer. What happens?
• Can you still parse it as integer somehow?
• Parse as double vec2 <- c('2', '3,45', '?', '-7,28')
• Parse correctly vec3 <- c('2', '3,45', '?', '-7.28')
• Parse the following guessing the parser: vec4 <- c('barrel: 432.7$', 'liter: 15.42PLN', 'gallon costs approx 32.1SEK', 'sunny, wind gusts up till 55m/s')
• Can you parse vec4 as number? Do it if you can.
• Parse vec5 <- "25 Dec 2015" as date (hint: ?parse_date()).
• Parse 10_Jul_1410 as date.

vec1 <- c(1, 7.2, 3.84, -5.23)
vec2 <- c('2', '3,45', '?', '-7,28')
vec3 <- c('2', '3,45', '?', '-7.28')
vec4 <- c('barrel: 432.7$', 'liter: 15.42PLN', 'gallon costs approx 32.1SEK', 'sunny, wind gusts up till 55m/s')
vec5 <- "25 Dec 2015"
parse_double(vec1)
parse_integer(vec1)
parse_integer(as.integer(vec1)) # Is it the best way? Hint: rounding.
parse_double(vec2, na = '?', locale = locale(decimal_mark = ','))
parse_number(vec2, na = '?', locale = locale(decimal_mark = ','))
guess_parser(vec4)
parse_guess(vec4)
# Yes, you can:
parse_number(vec4)
parse_date(vec5, format="%d %b %Y")
parse_date("10_Jul_1410", format="%d%.%b%.%Y")

3.1 Task 1 – selecting column

• load the nycflights13 package (install if necessary),
• read about the data in the package docs,
• inspect the flights tibble.
• select all columns but carrier and arr_time,
• select carrier, tailnum and origin,
• hide columns from day through carrier,
• select all columns that have to do with arrival (hint: ?tidyselect),
• select columns based on a vector v <- c("arr_time", "sched_arr_time", "arr_delay"),
• rename column dest to destination using:
  • select() and
  • rename()
    What is the difference between the two approaches?

install.packages('nycflights13')

library('nycflights13')

?nycflights13

flights

flights %>% select(-carrier, -arr_time)

flights %>% select(carrier, tailnum, origin)

flights %>% select(-(day:carrier))

flights %>% select(contains('arr_')) # or

v <- c("arr_time", "sched_arr_time", "arr_delay")
flights %>% select(v) # or
flights %>% select(one_of(v))

flights %>% select(destination = dest)
flights %>% rename(destination = dest)
# select keeps only the renamed column while rename returns the whole dataset
# with the column renamed

3.2 Task 2 – filtering rows

• filter only the flights that arrived ahead of schedule,
• filter the flights that had departure delay between 10 and 33,
• fish out all flights with unknown arrival time,
• retrieve rows 1234:1258 (hint: ?slice),
• sample (?sample_n()) 3 random flights per day in March,
• show 5 most departure-delayed flights in January per carrier,
• retrieve all unique() routes and sort them by origin,
• retrieve all distinct() routes and sort them by origin,
• is unique() more efficient than distinct()?

flights %>% filter(arr_delay < 0)

flights %>% filter(dep_delay >= 10, dep_delay <= 33) # or
flights %>% filter(between(dep_delay, 10, 33))

flights %>% filter(is.na(arr_time))

flights %>% slice(1234:1258)

flights %>% filter(month == 3) %>%
  group_by(day) %>%
  sample_n(3)

flights %>%
  filter(month == 1) %>%
  group_by(carrier) %>%
  top_n(5, dep_delay)

3.3 Task 3 – Trans(mutations)

• air_time is the amount of time in minutes spent in the air. Add a new column air_spd that will contain aircraft’s airspeed in mph,

• as above, but keep only the new air_spd variable,

• use rownames_to_column() on mtcars to add car model as an extra column,

flights %>% mutate(air_spd = distance/(air_time / 60))

flights %>% transmute(air_spd = distance/(air_time / 60))

mtcars %>% rownames_to_column('model')

3.4 Task 4 – groups and counts

• use group_by(), summarise() and n() to see how many planes were delayed (departure) every month,

flights %>%
  filter(dep_delay > 0) %>%
  group_by(month) %>%
  summarise(num_dep_delayed = n())

• do the same but using tally() and count(),

flights %>%
  filter(dep_delay > 0) %>%
  group_by(month) %>%
  tally()

flights %>%
  filter(dep_delay > 0) %>%
  count(month)

• what was the mean dep_delay per month?

flights %>%
  group_by(month) %>%
  summarise(mean_dep_delay = mean(dep_delay, na.rm = T))

• count the number of incoming delayed flights from each unique origin and sort origins by this count (descending),

flights %>%
  filter(arr_delay > 0) %>%
  group_by(origin) %>%
  summarise(cnt = n()) %>%
  arrange(desc(cnt))

• do the same using tally()

flights %>%
  filter(arr_delay > 0) %>%
  group_by(origin) %>%
  tally(sort = T)

• use summarise() to sum total dep_delay per month in hours,

 flights %>%
   group_by(month) %>%
   summarize(tot_dep_delay = sum(dep_delay/60, na.rm = T))

• use the wt parameter of count() (works with tally() too) to achieve the same,

 flights %>%
   group_by(month) %>%
   count(wt = dep_delay/60)

• run group_size() on carrier what does it return?

flights %>%
    group_by(carrier) %>%
    group_size()

• use n_groups() to check the number of unique origin-carrier pairs,

flights %>%
    group_by(carrier) %>%
    n_groups()

Note on ungroup Depending on the version of dplyr you may or may need to use the ungroup() if you want to group your data on some other variables:

flights %>%
  group_by(origin) %>%
  mutate(mean_delay_orig = (mean(dep_delay, na.rm = T) + mean(arr_delay, na.rm = T)) / 2) %>%
  ungroup() %>%
  group_by(carrier) %>%
  mutate(mean_delay_carr = (mean(dep_delay, na.rm = T) + mean(arr_delay, na.rm = T)) / 2) %>%
  select(origin, carrier, mean_delay_orig, mean_delay_carr)

may or may need ungroup depending on your dplyr version. In the newer versions, summarise and mutate drop one aggregation level.

3.5 Task 5 – joins

Given the following tibbles:

set1:

knitr::kable(set1)

and set2:

knitr::kable(set2)

Perform joins on id that result in the grey area from the Venn diagrams below:

left_join(set1, set2)

right_join(set1, set2)

inner_join(set1, set2) # or
semi_join(set1, set2) # semi_join removes duplicates in x
# and also returns only columns from x.

full_join(set1, set2) # or

anti_join(set1, set2) # or

5.1 Introduction

You will be given a fastq file coming from MinION sequencer (Oxford Nanopore). This file contains test reads from the chicken genome. The flow-cell used here has 512 channels, each channel consists of 4 pores and only one pore is active at a time. Once your sequence gets stuck for some reason, the device will attempt to remove it from the pore by playing with reversing polarity on that pore. If this was successful, the pore will be re-used. Your task will be to visualise reading events from the meta-data in the fastq dataset and to see how each and every channel behaved. Also, you will plot the distribution of reading times.

5.2 Preparations

First, we will need to load the necessary libraries. I will give you a hint – you need the following libraries:

• here – not necessary, but it is an elegant way of reading the data locally from the project folder,
• tidyverse – well, quite obvious why,
• ShortRead from Bioconductor – to deal with short reads in fastq,
• lubridate – to figure out reading times.

library(here)
library(tidyverse)
library(ShortRead)
library(lubridate)

5.3 Reading data

Now, let’s read the fastq data. Check ShortRead documentation to see how to read our fastq file. Also, try to use package here. If you write: data <- here::here('data/my.fastq'), the my.fastq file will be read from the data folder which is a subfolder of your project root, i.e. the folder where your r script is. It is a good practise and also prevents Jenny Brian from coming to your office and setting your computer on fire.

Now think a bit, to plot reading events, do we need al the data in the file or only some specific part? You may want to see some few first lines of the fastq to learn about the data structure.

raw_data <- here::here('docs/tidyverse_Marcin/lab/lab_assets/FUL1_fastqs_GRID2.fastq')
f <- FastqFile(raw_data)
rfq <- readFastq(f)
headers <- rfq@id
close(f)

5.4 Extracting information you need

In this step, we are extracting data from fastq headers of each and every read in the fastq file. Not super efficient and perhaps the slowest step of the whole analyses. Can you do it better than our example solution?

Desired output: a table (tibble/data.frame) with reads as rows and meta-data as columns.

Hint: use strsplit() to explode string data into columns and str_remove_all() to get rid of unwanted characters. Since we did not talk much about regular expressions and the stringr package:

• To split string on a particular character, group of characters use strsplit. Here we split on comma.

text <- "This text is long, or not?"
strsplit(text, ',')

• To remove everything following a given character, e.g. comma:

str_remove_all(text, ",.*")

data <- dplyr::as_tibble(matrix(NA_character_, ncol = 6, nrow = length(headers)))
colnames(data) <- c('id', 'run_id', 'sample_id', 'read', 'channel', 'start_time')
for (i in 1:length(headers)) {
  tmp <- toString(headers[[i]])
  tmp_flat <- unlist(strsplit(tmp, ' ')) %>%
    str_remove_all(".*=")
  data[i,] <- tmp_flat
}

5.5 Preparing tidy dataset

Now, the fun part begins:

• add column start_dttm that represents start time for a given read as proper datetime object (read lubridate docs) and
• add column chan that is the proper numeric representation of the channel,
• group reads by channel,
• arrange them by time,
• add time to next read (NA if this was the last read) and
• sort by channel again.

Hint: read about lead()

data2 <- data %>%
  mutate(start_dttm = as_datetime(start_time)) %>%
  mutate(chan=as.numeric(channel)) %>%
  group_by(chan) %>%
  arrange(start_dttm) %>%
  mutate(time_diff = lead(start_dttm) - start_dttm) %>%
  arrange(chan)

5.6 Plotting events per channel

Important! If you are not familiar with ggplot either:

• wait until tomorrow,
• use base-R plotting, or
• see our ggplot2 solution.

Here, we want to see what was happening in each channel over time. Plot the data you have just prepared so that:

• each point is the start of a new read,
• colour corresponds to the lag to the next read.

Can you visualise this in a better way? Different geometry?

ggplot(data2, mapping = aes(x = start_dttm,
                            y = as.factor(chan),
                            col = as.numeric(time_diff)
                            )
       ) +
  geom_point(size = .5) +
  theme_bw()

5.7 Distribution of time intervals

Now, we want to see how time-to-next-read is distributed. Since it has a veeeeery long right tail, I am cutting off everything above 200 seconds (just by eyeballing).

• plot time-to-next-read is distribution (you can use base-R histogram),
• can you find a better cutoff value?

# Show time-to-next read distribution
# thr <- mean(data2$time_diff, na.rm = T) + 3 * sd(data2$time_diff, na.rm = T)
tmp <- data2 %>%
  ungroup() %>%
  filter(time_diff < 200) %>%
  select(time_diff)

hist(as.numeric(tmp$time_diff), breaks = 1000, las=1)

6.1 Load necessary libraries

We will obviously need tidyverse, we will also do some string manipulations with stringr also here package is good to have.

library(tidyverse)
library(stringr)
library(here)

6.2 Input variables

Here, we will define our input variables. We need:

• file that contains the path to the dataset,
• sample_name is a string, the name of the sample you want to analyse,
• threshold is an intiger saying what is a the minimal number of replicates that have to contain an OTU in order to call it a true positive (TP),
• strict a logical. If set to TRUE, only the OTUs deemed TP will be shown.

Below we set some examle values:

# Change the path to your project path, where your data is
file <- here::here("docs/tidyverse_Marcin/lab/lab_assets/blast_result.csv")
sample_name <- 'SAMPLE12'
threshold <- 1
strict <- F

6.3 Reading the data

Now, you should read the data:

species_orig <- read_csv(file, col_names = c("sample","its","replicate","OTU","size","hit","perc_ident","score","family","species")) %>% select(-score)

As you see, the following information are included in the data:

• sample is simply the name of the sample,
• its is either ITS1 or ITS2 and tells which set of PCR primers has been used,
• replicate contains information on which replicate the sequences come from,
• OTU is a unique identifier of the so-called Operational Taxonomic Unit, an OTU often corresponds to one species but not always. Sometimes 2 OTUs represent the same species, sometimes 1 OTU consists of more than one species,
• size is the number of reads that support that particular OTU,
• hit is the BLAST hit identifier. The 4 top BLAST hits are reported per OTU,
• perc_identity is the percentage identity of the sequence to the BLAST hit,
• family is the identified plant family,
• species is the identified plant species.

6.4 Number of replicates per species

Create a new dataset species that contains an extra column n_replicates. The column contains number of replicates this particular species is present in. Do it per sample and its.

species <- species_orig %>%
  group_by(sample, its, species) %>%
  mutate(n_replicates = n_distinct(replicate)) %>%
  ungroup()

6.5 Filter out unwanted samples

Now, your task is to filter out all but your sample_name samples from the dataset. Call the resulting dataset my_sample.

my_sample <- species %>%
  filter(sample == sample_name)

6.6 Missing observations

What happens if a set of primers failed to amplify or if one replicate was lost? Use complete() to make sure you have NA values in such cases.

my_sample <- my_sample %>%
  complete(its = c("ITS1", "ITS2"),
           replicate = c("R1","R2","R3")
           )

6.7 Sorting issue

Look, the first sample in the table is SAMPLE10. Why not SAMPLE1? Thats a sorting issue: if sorted as character, 10 will come before 1. WE have to fix this by adding trailing zero to the values in OTU. We do not expect more than 99 OTUs in a sample, so it is ok with only one trailing 0 (otherwise the 100-th sample will spoil our sorting and come out like: SAMPLE100, SAMPLE01, SAMPLE10). We will need to use regular expression:

• all values in the OTU column that follow pattern “OTUdigit” we need to change to “OTU0digit”. Regular expression that matches this is OTU([0-9]$) and it should be replaced by: OTU0\\1. Ask your TAs to explain this if you do not know much about regular expressions and pattern matching.

my_sample <- my_sample %>% mutate(OTU = str_replace(OTU,
                                  pattern = "OTU([0-9]$)",
                                  replacement = "OTU0\\1"
                                )
                          )

6.8 Supporting reads

Sometimes, an OTU generates two or more top BLAST hits that come from the same species. We have decided to sum reads in such cases. Do it!

  my_sample <- my_sample %>%
    ungroup() %>%
    group_by(sample, its, replicate, OTU, species, n_replicates) %>%
    summarise(n_reads = sum(size)) %>%
    ungroup() %>%
    group_by(its, species, OTU)

6.9 Within-OTU species diversity

Now, we want to see how many identifications an OTU got. Implement this. Store the result in a new tibble diversity.

diversity <- my_sample %>%
  ungroup() %>%
  group_by(its, replicate, OTU) %>%
  summarise(n_species = n())

6.10 Adding diversity data

Add the diversity data to my_sample using appropriate join function. Also, remove the column with sample names since we are dealing with only one sample.

my_sample <- my_sample %>%
  left_join(diversity) %>%
  select(-sample)

• what columns did we join on in our example solution?

6.11 Visualising the data

Can you think of a good way of visualising the data? Think of:

• Type of plot you want. The simpler the better?
• Which variables you would like to visualise? We have chosen ITS, replicate, n_reads, n_replicates, OTU, threshold, n_species and specie. Well, pretty much all of them :-)
• How do you represent variables: colours, shapes, separate plots, sizes?
• What transformations do you need to apply before visualising the data?

Our example solution involves some ggplot2 magics. But would base-R be good enough for this type of plot?

Figure 6.1: Red – only one species in the current OTU, blue – identification above the threshold, grey – identification below the threshold