Introduction to the pedtools package

Magnus Dehli Vigeland

2020-08-04

The purpose of this vignette is to show how to work with pedigrees and marker data in pedtools.

The following command installs the current CRAN version of the package:

install.packages("pedtools")

Alternatively, you may want the latest development version from GitHub:

# install.packages("devtools") # install devtools if needed
devtools::install_github("magnusdv/pedtools")

Now you should be able to load pedtools.

library(pedtools)

1 Pedigrees

In pedtools and all related packages, pedigrees are stored as ped objects. We start by explaining briefly what these objects look like, and their basic constructor. If you are reading this vignette simply to learn how to create a particular pedigree, you may want to skip ahead to section 1.3 where we describe practical shortcuts to common pedigree structures.

1.1 The ped class

The ped constructor function
The most direct way to create a pedigree in pedtools is with the ped() constructor. This takes as input 4 vectors of equal length:

In other words, the j’th pedigree member has label id[j], father fid[j], mother mid[j], and gender given by sex[j].

For example, the following creates a family trio, i.e. father, mother and child:

ped(id = 1:3, fid = c(0,0,1), mid = c(0,0,2), sex = c(1,2,2))
#>  id fid mid sex
#>   1   *   *   1
#>   2   *   *   2
#>   3   1   2   2

In this example the child (id=3) is female, since the associated entry in sex is 2. Note that missing parents are printed as *. Individuals without parents are called founders of the pedigree, while the nonfounders have both parents specified. It is not allowed to have exactly one parent.

Instead of numerical labels as above, we could have used character strings. Let us create the trio again, with more informative labels, and store it in a variable named trio.

trio = ped(id = c("fa", "mo", "girl"), fid = c("","","fa"), mid = c("","","mo"), sex = c(1,2,2))
trio
#>    id fid mid sex
#>    fa   *   *   1
#>    mo   *   *   2
#>  girl  fa  mo   2

The special strings "0", "" and NA are all interpreted as a missing parent.

The internal structure of ped objects
From the way it is printed, the object trio appears to be a data frame, but this is not exactly true. Rather it is an object of class ped, which is basically a list. We can see the actual content of trio by unclassing it:

unclass(trio)
#> $ID
#> [1] "fa"   "mo"   "girl"
#> 
#> $FIDX
#> [1] 0 0 1
#> 
#> $MIDX
#> [1] 0 0 2
#> 
#> $SEX
#> [1] 1 2 2
#> 
#> $FAMID
#> [1] ""
#> 
#> $UNBROKEN_LOOPS
#> [1] FALSE
#> 
#> $LOOP_BREAKERS
#> NULL
#> 
#> $FOUNDER_INBREEDING
#> NULL
#> 
#> $MARKERS
#> NULL

In most cases it is not recommended for regular users to interact directly with the internal slots of a ped, since this can have unfortunate consequences unless you know exactly what you are doing. Instead, one should use accessor functions like labels(), getMarkers() and founderInbreeding(). The most important accessors are described within this vignette, while others are documented in the help page ?ped_utils.

1.2 Basic pedigree plots

To plot a pedigree, simply use plot().

plot(trio)

Under the hood, pedtools::plot() is an elaborate wrapper of the excellent plotting functionality of the kinship2 package. Most of the possibilities provided by kinship2 are available from pedtools, and several features are added. An overview can be found in the documentation ?plot.ped, but a quick example should get you started:

plot(trio, deceased = "fa", starred = "mo", hatched = "girl", 
     col = c("green", "red", "blue"), title = "Trio 1")

See Section 2.2 for how to add, and control the appearance of, marker genotypes to pedigree plots.

1.3 Built-in pedigree structures

Rather than using the ped() function directly, it is usually quicker and safer to build pedigrees step by step, applying the arsenal of utility functions offered by pedtools. A typical workflow is as follows:

  1. Choose one of the basic pedigree structures as starting point.
  2. Add/remove individuals as needed.
  3. Modify attributes like genders and labels.

You will find several examples below, but first let us list the available tools for each of the 3 steps.

Basic pedigrees
The following pedigree structures serve as starting points for pedigree constructions. For parameters and details, see ?ped_basic.

There are also more specialized structures, including double cousins, selfing pedigrees, and consecutive matings between full siblings. Look them up in ?ped_complex if you are interested.

Add/remove/extract individuals
The functions below are used to modify an existing ped object by adding/removing individuals, or extracting a sub-pedigree. For details, see ?ped_modify.

Edit labels and attributes
The following functions modify various attributes of a ped object. See ?ped_modify for parameters and details.

1.4 Examples of pedigree construction

Example 1: Trio

As our first example we will recreate the trio pedigree without using the ped() constructor. To give a hint of the flexibility, we show 3 alternative ways to code this.

Alternative A
The obvious starting point is nuclearPed(), with nch = 1 to indicate 1 child. By default, this creates a trio with numeric labels (father=1; mother=2; child=3) and a male child. Hence we fix the gender with swapSex(), and edit the labels with relabel():

trio2 = nuclearPed(nch = 1)
trio2 = swapSex(trio2, ids = 3)
trio2 = relabel(trio2, new = c("fa", "mo", "girl"))

Alternative B (quickest and best)
The previous approach can be condensed into a one-liner, since nuclearPed() allows an alternative syntax in which child genders and labels are specified directly:

trio3 = nuclearPed(father = "fa", mother = "mo", children = "girl", sex = 2)

Alternative C
Here is another possibility. We start by creating the father as a singleton, and then add the daughter:

trio4 = singleton("fa")
trio4 = addDaughter(trio4, parent = "fa", id = "girl")
#> Mother: Creating new individual with ID = NN_1
trio4 = relabel(trio4, old = "NN_1", new = "mo")

Note that addDaughter() automatically created the mother as “NN_1”, so we needed to relabel her.

Example 2: An inbred child

This time we will create this inbred family:

Alternative A
One approach is to first create individuals 1-6 as half sibships, with 1 child on the left side, and 2 children on the right. After this, we use addChildren() to add the inbred child.

x1 = halfSibPed(nch1 = 1, nch2 = 2, sex1 = 1, sex2 = 2:1)
x1 = addChildren(x1, father = 4, mother = 5, nch = 1)

Alternative B
We could also view the half siblings 4 and 5 as half cousins of degree 0. The halfCousinPed() function accepts an option child = TRUE adding an inbred child. The labels will be different with this approach, so you should plot the pedigree after each command to see who-is-who. Also, we must relabel in the end.

x2 = halfCousinPed(0, child = T)
x2 = addChildren(x2, father = 2, mother = 3, nch = 1)
x2 = relabel(x2, old = c(7,6), new = c(6,7))

A note about the order of pedigree members

Although both x1 and x2 reproduce exactly the plot shown above, they are not identical objects:

identical(x1, x2)
#> [1] FALSE

The reason is the order in which the individuals are stored. For x1 the ordering is the natural sequence 1,2,3,4,5,6,7, but for x2 our construction process has produced a slightly different order:

x2
#>  id fid mid sex
#>   1   *   *   2
#>   2   *   *   1
#>   3   *   *   2
#>   4   2   1   1
#>   5   2   3   2
#>   7   4   5   1
#>   6   2   3   1

The internal ordering is usually of little importance in applications.1 However, if you get annoyed by “wrong” orderings such as for x2 above, you can use reorderPed() to permute the pedigree any way you like. In fact, the default action of this function is to permute into the natural order of the labels, which is exactly what we need to make x2 identical to x1:

x2 = reorderPed(x2)
identical(x1, x2)
#> [1] TRUE

1.4.1 Example 3: A complex family tree

For our final example we consider a complicated family tree extending both upwards and downwards from a single person.

We will use this example to demonstrate the mergePed() function. When this function is given two pedigrees, it “glues together” members with matching ID labels, and checks that the result is a valid pedigree.

The hardest part of using mergePed() is to get the labelling right; this will almost always involve the relabel() function. To keep track of the labels, you should plot after each new line of code. Here is how the pedigree was created:

# Top part
x = ancestralPed(g = 2) # bottom person is `7`; relabel to `8`
x = relabel(x, old = 7, new = 8)

# Bottom part
y = halfCousinPed(degree = 1) 
y = swapSex(y, 4)
#> Changing sex of spouses as well: 5
y = relabel(y, new = 7:15) # top person becomes `8`

# Merge
z = mergePed(x, y)

1.5 Pedigree subsets

Pedtools offers a range of utility functions for identifying subsets of pedigree members. These come in two flavours: 1) members with certain global property, and 2) members with a certain relationship to a given individual.

Pedigree members with a certain property
Each of the following functions returns a vector specifying the members with the given property.

By default, the output of these functions is a character vector containing ID labels. However, adding the option internal = TRUE will give you an integer vector instead, reporting the internal indices of the members. This is frequently used in the source code of pedtools, but is usually not intended for end users of the package.

Relatives of a given individual
The functions below take as input a ped object and the label of a single member. They return a vector of all members with the given relation to that individual.

2 Markers

The other main theme of the pedtools package (pedigrees being the first) are marker genotypes.

2.1 Creating marker objects

Marker objects created with the marker() function. For example, the following command makes an empty marker associated with the trio pedigree:

marker(trio)
#>    id  <NA>
#>    fa   -/-
#>    mo   -/-
#>  girl   -/-
#> * * * * * *  
#> Position: NA 
#> Mutation: none 
#> Frequencies:
#>    1   2
#>  0.5 0.5

As shown in the output, the marker is indeed empty: All pedigree members have missing genotypes, and there is no assigned name or position. By default, markers are diallelic, with alleles 1 and 2, with equal frequencies. For a more interesting example, let us make a SNP named “snp1”, with alleles “A” and “B”. The father is homozygous “A/A”, while the mother is heterozygous. We store it in a variable m1 for later use.

m1 = marker(trio, fa = "A/A", mo = "A/B", name = "snp1")

This illustrates several points. Firstly, individual genotypes may be specified using the ID labels. The different alleles occurring in the genotypes is interpreted as the complete set of alleles for the marker. Finally, these are assigned equal frequencies. Of course, this behaviour can be overridden, by declaring alleles frequencies explicitly:

marker(trio, fa = "A/A", mo = "A/B", afreq = c(A = .2, B = .3, C = .5))
#>    id  <NA>
#>    fa   A/A
#>    mo   A/B
#>  girl   -/-
#> * * * * * *  
#> Position: NA 
#> Mutation: none 
#> Frequencies:
#>    A   B   C
#>  0.2 0.3 0.5

The markers chromosome can be declared using the chrom argument, and similarly its position by posMb (megabases). Markers with unknown chromosome are treated as autosomal. To define an X-linked marker, put chrom = "X". the fact that males are hemizygous on X (i.e. they have only one allele) is reflected in the printout of such markers:

m2 = marker(trio, fa = "A/A", mo = "A/B", chrom = "X", name = "snpX")
m2
#>    id  snpX
#>    fa     A
#>    mo   A/B
#>  girl   -/-
#> * * * * * *  
#> Position: chr = X, Mb = NA 
#> Mutation: none 
#> Frequencies:
#>    A   B
#>  0.5 0.5

A side note: It may come as a surprise that you don’t need quotes around the ID labels (which are characters!) in the above commands. This is because marker() uses non-standard evaluation (NSE), a peculiarity of the R language which often leads to less typing and more readable code.2 Unfortunately, this doesn’t work with numerical ID labels. Thus to assign a genotype to someone labelled “1” you need quotes, as in marker(trio, "1" = "A/A").

2.2 Plotting pedigrees with marker data

Including marker data in a pedigree plot is straightforward:

plot(trio, marker = m1)

The appearance of the genotypes can be tweaked in various ways, as documented in ?plot.ped. Here’s an example:

plot(trio, marker = m1, showEmpty = T, missing = "?", sep = "")

2.3 Markers attached to ped objects

Although a ped object is needed in the creation of a marker, the two are independent of each other once the marker is created. In many applications it is useful to attach markers to their ped object. In particular for bigger projects with many markers, this makes it easier to manipulate the dataset as a unit.

To attach a marker m (which could be a list of several markers) to a pedigree x, there are two options:

The difference between these is that setMarkers() replaces all existing markers, while addMarkers() appends m to the existing ones. In our trio example the two are equivalent since there are no existing markers.

trio = setMarkers(trio, list(m1, m2))
trio
#>    id fid mid sex snp1 snpX
#>    fa   *   *   1  A/A    A
#>    mo   *   *   2  A/B  A/B
#>  girl  fa  mo   2  -/-  -/-

Selecting and removing attached markers
Four closely related functions functions are useful for manipulating markers attached to a pedigree:

All of these have exactly the same arguments, described in more detail in ?marker_select. Let us do a couple of examples here. Recall that by now, our trio has two attached markers; the first is called “snp1”, and the other is on the X chromosome.

whichMarkers(trio, chrom = "X")
#> [1] 2
selectMarkers(trio, markers = "snp1")
#>    id fid mid sex snp1
#>    fa   *   *   1  A/A
#>    mo   *   *   2  A/B
#>  girl  fa  mo   2  -/-

2.4 Accessing and modifying individual markers

Internally, a marker object is stored as a matrix with two columns (one for each allele) and one row for each pedigree member. The matrix is numeric (for computational convenience) while the allele labels and other meta information are added as attributes. The most important of these are:

In addition to those listed above, there are two more attributes: pedmembers and sex. They store the ID labels and genders of the pedigree associated with the marker, and are only used to empower the printing method of marker objects.

Marker accessor functions
For each marker attribute listed above, there is a corresponding function with the same name for retrieving its content. These functions take as input either a marker object, or a ped object together with the name (or index) of an attached marker. This may sound a bit confusing, but a few examples will make it clear!

Recall that our marker “snp1” exists in two copies: One is stored in the variable m1, while the other is attached to trio. In both cases we can extract the allele frequencies with the function afreq().

afreq(m1)
#>   A   B 
#> 0.5 0.5
afreq(trio, marker = "snp1")
#>   A   B 
#> 0.5 0.5

We can also modify the frequencies using this syntax. To avoid confusion about the allele order, the frequencies must be named with the allele labels (just as in the output of afreq() above).

afreq(trio, marker = "snp1") = c(A = 0.9, B = 0.1)

In addition to the functions getting and setting marker attributes, there is one more important marker accessor, namely genotype(). This returns the genotype of a specified individual, and can also be used to modify genotypes. As the others, it can be applied to marker objects directly, or to pedigrees with attached markers. Here we show a few examples of the latter type:

# Girl is not genotyped
genotype(trio, "snpX", id = "girl")
#> [1] NA NA

# Set genotype
genotype(trio, "snpX", id = "girl") = "A/A"

# Inspect the result
trio
#>    id fid mid sex snp1 snpX
#>    fa   *   *   1  A/A    A
#>    mo   *   *   2  A/B  A/B
#>  girl  fa  mo   2  -/-  A/A

2.5 Getting/setting/modifying many markers simultaneously

Several functions in pedtools are indented for modifying many (or all) markers at the same time. Their purpose and typical use cases are summarised in the table below. The argument x always denotes a ped object.
Use … When you want to … For example to …
Get
getAlleles(x) extract all alleles as a matrix. do summary stats on the marker alleles
getFreqDatabase(x) extract allele frequencies as a data.frame in allelic ladder format. transfer to other objects, or write the database to a file
getMarkers(x) extract list of marker objects. Each marker is a N * 2 allele matrix (N = pedsize(x)) with locus annotations as attributes do computations
Set
setAlleles(x, ...) replace the genotypes of x without changing the locus attributes. erase all genotypes
setFreqDatabase(x, db) replace all allele frequencies without changing the genotype data. The input is a data.frame in allelic ladder format. Conceptually equivalent to setMarkers(x, alleleMatrix = getAlleles(x), locusAnnotations = db). change the frequency database
setMarkers(x, ...) attach marker objects with or without genotype data. Locus attributes are indicated as a list; genotypes as a matrix or data.frame. prepare joint manipulation of a pedigree and marker data
Convert
as.data.frame(x) convert x to a data.frame, with pedigree columns in standard format followed by genotype columns. One column per marker, with genotype format a/b and missing alleles indicated as -. pretty-print ped objects
as.matrix(x) convert x to a numerical matrix, with additional info attached as attributes. modify a pedigree with marker data
Transfer
transferMarkers(from, to) transfer genotypes and attributes between pedigree objects (or lists of such). transfer simulated marker data

  1. There is an important exception to this: Certain algorithms in pedigree analysis work “top-down”, in the sense that parents must be treated before their children. For this reason, many implementations require, for simplicity, that the individuals are stored in this fashion, i.e. that parents always precede their children. pedtools offers a special reordering function to ensure this, parents_before_children(), which you will find utilised in the source code of packages like ribd and ibdsim2.↩︎

  2. You may have come across NSE before, for instance when using subset() on a data.frame. To learn more about NSE, I recommend this book chapter by Hadley Wickham:
    http://adv-r.had.co.nz/Computing-on-the-language.html↩︎