The purpose of this vignette is to show how to work with pedigrees and marker data in
The following command installs the current CRAN version of the package:
Alternatively, you may want the latest development version from GitHub:
Now you should be able to load
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.
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:
id: individual ID labels (numeric or character)
fid: id of the fathers (0 if not included)
mid: id of the mothers (0 if not included)
sex: gender codes, with entries 0 (unknown), 1 (male) or 2 (female)
In other words, the j’th pedigree member has label
mid[j], and gender given by
For example, the following creates a family trio, i.e. father, mother and child:
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
The special strings
NA are all interpreted as a missing parent.
The internal structure of
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:
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
founderInbreeding(). The most important accessors are described within this vignette, while others are documented in the help page
To plot a pedigree, simply use
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:
See Section 2.2 for how to add, and control the appearance of, marker genotypes to pedigree plots.
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:
You will find several examples below, but first let us list the available tools for each of the 3 steps.
The following pedigree structures serve as starting points for pedigree constructions. For parameters and details, see
singleton(), a pedigree consisting of a single individual
nuclearPed(), a nuclear pedigree (parents+children)
halfSibPed(), two sibships with one parent in common
linearPed(), a straight line of successors
cousinPed(), cousins of specified degree/removal
halfCousinPed(), half cousins of specified degree/removal
ancestralPed(), a family tree containing the ancestors of a single person
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.
The functions below are used to modify an existing
ped object by adding/removing individuals, or extracting a sub-pedigree. For details, see
addChildren(), with special cases
Edit labels and attributes
The following functions modify various attributes of a
ped object. See
?ped_modify for parameters and details.
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.
The obvious starting point is
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
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:
Here is another possibility. We start by creating the father as a singleton, and then add the daughter:
addDaughter() automatically created the mother as “NN_1”, so we needed to relabel her.
This time we will create this inbred family:
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.
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.
A note about the order of pedigree members
x2 reproduce exactly the plot shown above, they are not identical objects:
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:
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
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:
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.
The other main theme of the
pedtools package (pedigrees being the first) are marker genotypes.
Marker objects created with the
marker() function. For example, the following command makes an empty marker associated with the
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.
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:
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:
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").
Including marker data in a pedigree plot is straightforward:
The appearance of the genotypes can be tweaked in various ways, as documented in
?plot.ped. Here’s an example:
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
m to the existing ones. In our
trio example the two are equivalent since there are no existing markers.
Selecting and removing attached markers
Four closely related functions functions are useful for manipulating markers attached to a pedigree:
selectMarkers(), returns a
pedobject where only the indicated markers are retained
removeMarkers(), returns a
pedobject where the indicated markers are removed
getMarkers(), returns a list of the indicated markers
whichMarkers(), returns the indices of the indicated markers
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.
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:
alleles: The allele labels, stored as a character vector.
afreq: The allele frequencies, in the same order as the alleles. An error is issued if the frequencies do not sum to 1 after rounding to 3 decimals.
name: The marker name, which can be any character string not consisting solely of digits.
chrom: The chromosome name. This can be given as an integer, but is always converted to character. The special values “23” and “X” are recognized as the human X chromosome, which affects the way genotypes are printed.
posMb: Chromosomal position given in megabases.
In addition to those listed above, there are two more attributes:
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
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
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:
pedtoolsare 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
xalways denotes a
|Use …||When you want to …||For example to …|
||extract all alleles as a matrix.||do summary stats on the marker alleles|
||extract allele frequencies as a data.frame in allelic ladder format.||transfer to other objects, or write the database to a file|
extract list of marker objects. Each marker is a
replace the genotypes of
||erase all genotypes|
replace all allele frequencies without changing the genotype data. The input is a data.frame in allelic ladder format. Conceptually equivalent to
||change the frequency database|
||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|
||pretty-print ped objects|
||modify a pedigree with marker data|
||transfer genotypes and attributes between pedigree objects (or lists of such).||transfer simulated marker data|
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
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: