Object-Oriented Programming in R

Note

During this lab, we will go through the most important features of 3 out of the 4 existing OOP systems in R:

While S3 and S4 are base-R OOP systems, R6 is provided by the R6 package. Both S3 and S4 follow the functional OOP style while R6 is more similar to OOP models known from Java or C++. We will not cover the Reference Classes (a.k.a. RC) which is yet another base-R OOP model. Why? Well, it is really tricky to use and its documentation is not always complete…

1 S3 Classes

What is the class of the object returned by the lm() function?

Code

obj <- lm(speed ~ dist, data = cars) 
class(obj)

What basic data type is it build upon?

Code

typeof(obj)

What attributes does the object of the lm class have?

Code

attributes(obj)

What is the structure of the lm object?

Code

str(obj)

Does the lm class implement own str()?

Code

str(obj)
# compare to:
str(unclass(obj)) 

# Answer: no

What is the class of a tibble? You have to load the tidyverse package and use on of the built-in datasets, e.g. mtcars as argument to as_tibble() to check the class.

Code

require(tidyverse)
obj <- as_tibble(mtcars)
class(obj)

What is the underlying data type?

Code

typeof(obj)

Is the str() function used by tibbles the default str()?

Code

str(obj)
str(unclass(obj))
# Answer: no

Implement your own class meta_frame that is a data.frame with additional attributes:
- Character attribute descr (a description),
- creation attribute which will hold the creation date.

Check that it works, i.e. that it has the expected attributes and that it is a separate class.

How do you access the description field?

Code

meta_frame <- function(df, descr, creation) {
  structure(
    list(
      df = df, 
      descr = descr, 
      creation = creation
    ),
    class = "meta_frame"
  )
}
obj <- meta_frame(mtcars, 'The mtcars dataset', '10-06-2018')
str(obj)
obj$descr

Design implementation of your own S3 class protein that stores the following information (you will implement the class in the next point, now just think and draw on paper if you wish):
- Protein sequence,
- Protein length,
- Post-translational modification (PTM) site coords,
- PTM type that can be either of: “phosphorylation”, or “methylation”.
Implement the protein class from the previous point using the constructor, validator and helper design pattern. Try to reduce the number of input arguments to: + + a sequence and
- A list of PTMs.

Code

# Constructor
new_protein <- function(sequence, length, ptm_site, ptm_type) {
  stopifnot(is.character(sequence))
  stopifnot(is.integer(length))
  stopifnot(is.numeric(ptm_site))
  stopifnot(is.character(ptm_type))
  structure(
    sequence,
    length = length,
    ptm_site = ptm_site,
    ptm_type = ptm_type,
    class = "protein"
  )
}

# Helper
protein <- function(sequence, ptm) {
  sequence <- sequence
  len <- nchar(sequence)
  ptm_site <- ptm[[1]]
  ptm_type = ptm[[2]]
  validate_protein(new_protein(sequence, len, ptm_site, ptm_type))
}

# Validator
validate_protein <- function(x) {
  sequence <- x[1]
  len <- attr(x, 'length')
  ptm_site <- attr(x, 'ptm_site')
  ptm_type <- attr(x, 'ptm_type')
  if (is.na(sequence)) {
    stop('Sequence is missing!', call. = FALSE)
  }
  if (length(ptm_site) != length(ptm_type)) {
    stop('PTM type and site differ in length!', call. = FALSE)
  }
  if ((sum(!ptm_type %in% c('phosphorylation', 'methylation'))) > 0) {
    stop('Invalid PTM types detected!', call. = FALSE)
  }
  return(x)
}

What would you improve in our implementation of protein. Is it really user friendly?
Write a generic print for the protein class. It should print e.g.:

[1] "============== Protein =============="
[1] "Sequence:ARNDKLLQWYTTARD"
[1] "Length: 15 aa."
[1] "============== PTM section =============="
[1] "Site: 3" "Site: 5"
[1] "Type: phosphorylation" "Type: methylation"

Code

print.protein <- function(x) {
  sequence <- unclass(x)
  len <- attr(x, 'length')
  ptm_site <- attr(x, 'ptm_site')
  ptm_type <- attr(x, 'ptm_type')
  print("============== Protein ==============")
  print(paste0("Sequence:", sequence))
  print(paste0("Length: ", len, " aa."))
  print("============== PTM section ==============")
  print(paste0("Site: ", ptm_site))
  print(paste0("Type: ", ptm_type))
}

Test that the protein class works as it should and that generic print works as well.

Code

my_prot <- protein("ARNDKLLQWYTTARD", 
                   list(
                     site = c(3, 5), 
                     type = c('phosphorylation', 'methylation')
                    )
                  )
class(my_prot)
typeof(my_prot)
print(my_prot)

2 S4 Classes

Re-write the protein class in S4 (with validation!).

Code

# Generator
.protein <- setClass('protein',
                     slots = c(
                       sequence = 'character',
                       length = 'numeric',
                       ptm_site = 'numeric',
                       ptm_type = 'character'
                      )
                    )

# Constructor
protein <- function(sequence, ptm) {
  sequence <- sequence
  len <- nchar(sequence)
  ptm_site <- ptm[[1]]
  ptm_type <- ptm[[2]]
  if (is.na(sequence)) {
    stop('Sequence is missing!', call. = FALSE)
  }
  if (length(ptm_site) != length(ptm_type)) {
    stop('PTM type and site differ in length!', call. = FALSE)
  }
  if ((sum(!ptm_type %in% c('phosphorylation', 'methylation'))) > 0) {
    stop('Invalid PTM types detected!', call. = FALSE)
  }
  pt <- .protein(sequence = sequence, 
           length = len, 
           ptm_site = ptm_site, 
           ptm_type = ptm_type)
  return(pt)
}

Create an S4 object of the protein class and check whether it works.

Code

my_prot <- protein("ARNDKLLQWYTTARD", 
                   list(
                     site = c(3, 5), 
                     type = c('phosphorylation', 'methylation')
                    )
                  )
class(my_prot)
typeof(my_prot)
str(my_prot)

Implement the generic print using S4 and check that it works.

Code

setMethod('print', 'protein', 
  function(x) {
    sequence <- x@sequence
    len <- x@length
    ptm_site <- x@ptm_site
    ptm_type <- x@ptm_type
    print("============== Protein ==============")
    print(paste0("Sequence:", sequence))
    print(paste0("Length: ", len, " aa."))
    print("============== PTM section ==============")
    print(paste0("Site: ", ptm_site))
    print(paste0("Type: ", ptm_type))
})

print(my_prot)

Implement a new S4 class ext_protein that extends protein with 3 slots:
- Feature type,
- Feature position, and
- Feature value.

Code

.ext_protein <- setClass('ext_protein', 
                         contains = c('protein'),
                         slots = c(
                           prot = 'protein',
                           feature_type = 'character',
                           feature_position = 'numeric',
                           feature_value = 'character'
                         ))
my_ext_prot <- .ext_protein(prot = my_prot,
                           feature_type = 'modification',
                           feature_position = 11,
                           feature_value = 'absent'
                  )
class(my_ext_prot)
typeof(my_ext_prot)
str(my_ext_prot)

3 R6 Classes

Install and load the R6 package,
Implement the protein class using R6 model and check that it works as expected:

Code

require(R6)
protein <- R6Class(classname = 'protein',
                        public = list(
                          seq = NA,
                          length = NULL,
                          ptm_site = NA,
                          ptm_type = NA,
                          initialize = function(seq = NA, ptm = NA) {
                            self$seq <- seq
                            self$length <- nchar(self$seq)
                            self$ptm_site <- ptm[[1]]
                            self$ptm_type <- ptm[[2]]
                            # Check types
                            stopifnot(is.character(seq))
                            
                            #Validate
                            if (is.na(self$seq)) {
                              stop('Sequence is missing!', call. = FALSE)
                            }
                            if (length(self$ptm_site) != length(self$ptm_type)) {
                              stop('PTM type and site differ in length!', call. = FALSE)
                            }
                            if ((sum(!self$ptm_type %in% c('phosphorylation', 'methylation'))) > 0) {
    stop('Invalid PTM types detected!', call. = FALSE)
  }
                          }
 )
)

my_new_prot <- protein$new(seq = "ARNDKLLQWYTTARD", ptm = 
                   list(
                     site = c(3, 5), 
                     type = c('phosphorylation', 'methylation')
                    ))
str(my_new_prot)

Congratulations! You are familiar with S3, S4 and R6 object models by now!