Introduction to dartR

Bernd Gruber, Peter Unmack, Oliver Berry & Arthur Georges

2020-09-01

Contents
1 R installation 4
1.1 Installing dartR from CRAN . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2 Manual installation via Github (with all dependencies) . . . . . . . . . . . . . . . . . . . . . . 5
1.3 Testing the installation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.4 Setting the working directory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.5 Test datasets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

2 Genlight Format 6

3 DArT Input Data Formats 8

4 Reading DArT Files into a Genlight Object 9

5 Working with DArT Genlight Objects 11

5.1 Locus metadata . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
5.2 Individual Metadata . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
5.3 Importing data from other formats/sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

1
6 Subsetting and Recoding Data 21
6.1 Filtering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
6.2 Examples of dartR code to filter a gl dataset . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

7 Subsetting and Recoding Data 24

7.1 Population (=higher level grouping) reassignment . . . . . . . . . . . . . . . . . . . . . . . . . 24
7.2 Individual reassignment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
7.3 Recode tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
7.4 Deleting individuals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
7.5 Recalculating locus metadata . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
7.6 Using R commands to manipulate the genlight object . . . . . . . . . . . . . . . . . . . . . . 31

8 Genetic Distance 33
8.1 F Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

9 Visualisation 39
9.1 PCoA in dartR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
9.2 Plotting the results of PCoA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
9.3 The Scree Plot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
9.4 3D Plot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
9.5 Neighbour-joining trees . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48

10 Fixed Difference Analysis 49

11 Isolation by distance plot 51

12 Phylogenetic Analysis 52
12.1 Distance Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
12.2 Character-based Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
12.3 Converting Diploid to Haploid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
12.4 Using Ambiguity Codes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57

13 Population Assignment 60

14 Sex Linkage 63

15 Conversion to formats for other packages 64

15.1 Interfaces to Other Software . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
15.2 NewHybrids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
15.3 Phylip . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
15.4 SNPRelate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
15.5 FastSTRUCTURE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66

2
16 References 66

3
#What is dartR?
Package dartR is an R package for a) loading DArT TM SNP and SilicoDArT data generated from the
commercial service provided by Diversity Arrays Technology Pty Ltd; (b) applying filters to those data
based on locus metadata such as call rate, information content or reproducibility; (c) assigning individuals to
populations and selecting subsets of individuals or populations; (d) visualization using Principal Coordinates
Analysis (PCoA), (e) initial calculation of indices such as heterozygosity and Fst ; and (f) providing a conduit
to a range of standard data formats and R packages for analysis.
Provided the data are in the genlight format (package adegenet), the package can be used to analyse SNP
data from sources other than DArT TM . Please note we provide give a detailed example below and outline
several ways how to import data from other formats into genlight.
In most cases, the scripts in {dartR} are wrappers for scripts included in other already available packages,
to provide easy access to these packages for analyzing DArT data, and to provide some enhanced output
diagnostics. Relatively few scripts provide novel analyses. We make no apologies for this, as the objective of
dartR is to provide fundamental tools for accessing and manipulating SNP data in preparation for analysis
by the vast suite of packages available in R through the CRAN repository, and make them easily accessible
for users not deeply accustomed with the R language, without “re-inventing the wheeel”.
A summary of the capabilities of dartR is as follows:

• Intelligent interpretation and input of DArT comma-delimited files to a compact genlight form of the
R {adegenet} package.
• Filtering loci and individuals on criteria drawn from the DArT locus metadata (such as repAvg, Avg-
PIC) or on computed statistics (such as call rate or Hamming distance).
• Relabelling individuals and recoding populations into new aggregations, and deleting selected individ-
uals or populations.
• Visualization using Principal Coordinates Analysis (PCoA), Neighbour-joining trees and Isolation-by-
distance analysis.
• Translation to other R packages (e.g. NewHybrids), to other {adegenet} objects (e.g. genind), and to
standard data formats (e.g. fastA).
• A few specific analyses not available elsewhere (e.g. fixed difference analysis, assignment testing, fast
calculation or linkage disquilibria per population).

1 R installation
This tutorial will not provide information how to use R. For that please search the web using the keywords
“R tutorial introduction beginner” and you will find many excellent resources. To be able to run the code
in this tutorial you first need to download and install R via https://www.r-project.org/. Once installed we
strongly recommend to download and install R-studio via https://www.rstudio.com/, which is becoming the
defacto GUI for R. When using different packages you may get warning messages about packages built under
different versions of R, these warnings can generally be ignored. If you are unfamiliar with using R there are
a few important issues to be aware of. Commands and file names are case sensitive, when providing a path
only use single forward slashes “/” (even if your operating system uses back slashes).
#Using this Vignette
To use this vignette interactively, you need to install and load the dartR package, and set the default
directory using setwd(). To install the latest developmental version of the dartR package visit the github
page for instructions (https://github.com/green-striped-gecko/dartR). Then we make the package available
into the current project via library(dartR). A brief description on the installation procedure is given below.
There are two ways to install dartR, either via CRAN (prefered for the standard user) or via Github (latest,
but untested version).

4
1.1 Installing dartR from CRAN

The latest stable version of dartR can be installed from the CRAN repository. Simply type:

install.packages("dartR")

1.2 Manual installation via Github (with all dependencies)

install.packages("devtools")
library(devtools)
install.packages("BiocManager")
BiocManager::install(c("SNPRelate", "qvalue"))
install_github("green-striped-gecko/dartR")
library(dartR)

1.3 Testing the installation

To test if your installation was successful include the library by typing:

library(dartR)

## Loading required package: adegenet

## Loading required package: ade4

## Registered S3 method overwritten by ’spdep’:

## method from
## plot.mst ape

##
## /// adegenet 2.1.3 is loaded ////////////
##
## > overview: ’?adegenet’
## > tutorials/doc/questions: ’adegenetWeb()’
## > bug reports/feature requests: adegenetIssues()

## Loading required package: ggplot2

## Registered S3 method overwritten by ’pegas’:

## method from
## print.amova ade4

## Registered S3 method overwritten by ’GGally’:

## method from
## +.gg ggplot2

## Registered S3 method overwritten by ’genetics’:

## method from
## [.haplotype pegas

5
If some packages have not been loaded, an error message will be given and you will need to install those
package manually (see a detailed description of the process at the Github repository. A common problem is
also that a package from bioiconductor is not installed (SNPRelate or qvalue). To fix this problem type:

BiocManager::install(c("SNPRelate", "qvalue"))

1.4 Setting the working directory

Once installed and invoked into your current R session, we need to prepare our session, so it points to a
working directory. The working directory is the location on your hard drive, that holds your data and also
is where intermediate and final output files will be exported if not directed otherwise.
To set the default directory, use the following with the appropriate directory specified:

# Set the default working directory (change this to suit)

setwd("c:/your.working.directory/")

Note that the file specification uses forward slashes, irrespective of which operating system you use (Linux,
Windows or MacOS).

1.5 Test datasets

The dartR library contains test datasets that form the basis of the commands and exercises below. We
recommend that you rename the test data set to gl in case you want to make alterations to it. If you wish
to run the vignette code with your own data, simply rename your data set to gl after you have loaded it
into R.

# Rename the test genlight object to gl, something simple

gl <- testset.gl

When working through the vignette, you should try the commands to see if you can replicate the output.

2 Genlight Format
The R package dartR relies on the SNP data being stored in a compact form using a bit-level coding scheme.
SNP data coded in this way are held in a genlight object that is defined in R package adegenet (Jombart,
2008; Jombart and Ahmed, 2011). Refer to the tutorial prepared by Jombart and Collinson (2015) called
Analysing genome-wide SNP data using adegenet 2.0.0, if you require further information.
The complex storage arrangement of genlight objects is hidden from the user because it is accompanied by
a number of accessors. These allow the data to be accessed in a way similar to the manipulation of standard
objects in R, such as lists, vectors and matrices.
A genlight object can be considered to be a matrix containing the SNP data encoded in a particular way. The
matrix entities (rows) are the individuals, and the attributes (columns) are the SNP loci. In the body of this
individual x locus matrix are the SNP data, coded as 0 for homozygous reference state, 1 for heterozygous,
and 2 for homozygous alternate (or SNP) state. You can access these data by converting to a standard
matrix using

6
m <- as.matrix(gl)

This function allows normal R approaches for examination and manipulation. For example,

as.matrix(gl)[1:5,1:3]

## 100049687-12-C/T 100049698-16-G/A 100049728-23-A/G

## AA010915 2 NA 0
## UC_00126 2 NA 0
## AA032760 NA NA 0
## AA013214 2 NA 0
## AA011723 2 NA 0

displays the SNP states for 5 rows [individuals] and 3 columns [loci]. Associated with the SNP genotypes in
the genlight object is a data.frame of locus metadata (e.g. CloneID, CallRate, TrimmedSequence, etc). Each
item in this list (loc.metrics) is a vector of length equal to the number of loci. Also associated with the SNP
genotypes is a data.frame of individual metadata (individual id, population, sex, lat(itude), lon(gitude),
etc) Each item in this data.frame (ind.metrics) is a vector of length equal to the number of individuals.
The information in the genlight object can be accessed using the following adegenet accessors:

• nInd(gl): | returns the number of individuals in the genlight object.

• nLoc(gl): returns the number of loci.

• nPop(gl): returns the number of populations to which the individuals are assigned.
• indNames(gl): returns or sets labels for individuals.
• locNames(gl): returns or sets labels for loci.

• alleles(gl): returns or sets allelic states of each locus for each individual (e.g. “A/T”).
• ploidy(gl): returns or sets the ploidy of the individuals (normally diploid or 2).
• pop(gl): returns or sets the population to which each individual belongs. Try also levels(pop(gl)) for
a list of unique population names.

• NA.posi: returns the loci with missing values, that is, loci for which a sequence tag failed to amplify
for each individual.
• chr: returns or sets the chromosome for each locus.
• position: returns or sets the position of each SNP in the sequence tag of each locus.

• other(gl): returns of sets miscellaneous information stored as a list.

Task
Try some of these using commands on the R console

An alternative way to write these accessors is to use the form of gl@pop or gl@other, for example.

7
Some simple operations such as computing allele frequencies or diagnosing missing values can be problematic
without representing the full dataset in memory. The package {adegenet} implements a few procedures that
perform such basic tasks on genlight objects, processing one individual at a time, thereby minimizing memory
requirements.

• glSum: counts the frequency of the alternate allele for each locus.
• glNA: counts the number of missing values for each locus.
• glMean: computes the relative frequency (a proportion in the range 0-1) of the second allele for each
locus.
• glVar: computes the variance of the allele frequency distribution for each locus.
• glDotProd: computes the dot products between all pairs of individuals, with centering and scaling.

in each case taking advantage of the 0, 1, 2 coding of the SNP states for each locus - the coding essentially
corresponds to the frequency of the alternate allele.

Task
Try some of these using commands on the R console

Hint
For further information on using genlight objects for SNP datasets, refer to the tutorial prepared
by Thibaut Jombart and Caitlin Collinson (2015) entitled Analysing genome-wide SNP data using
adegenet 2.0.0.

3 DArT Input Data Formats

Diversity Arrays Technology Pty Ltd (DArT™) supply your data as excel spreadsheets in comma delimited
format (.csv). Several files are provided.

• SNP_1row.csv contains the SNP genotypes in one row format

• SNP_2row.csv contains the SNP genotypes in two row format
• SilicoDArT.csv contains the presence(1)/absence(0) of the sequence tag at a locus for each individual
(analogous to AFLPs)
• metadata.csv contains a report of the success of the sequencing and an explanation of the locus
metadata provided in the above spreadsheets.

Hint
Refer to the DArT documentation provided with your report for further information.

8
4 Reading DArT Files into a Genlight Object
SNP data can be read into a genlight object using read.dart(). This function intelligently interrogates the
input csv file to determine * if the file is a 1-row or 2-row format, as supplied by Diversity Arrays Technology
Pty Ltd. * the number of locus metadata columns to be input (the first typically being cloneID and the last
repAvg). * the number of lines to skip at the top of the csv file before reading the specimen IDs and then
the SNP data themselves. * if there are any errors in the data. An example of the function used to input
data is as follows:

gl <- gl.read.dart(filename = "testset.csv", ind.metafile = " ind_metrics.csv")

The filename specifies the csv file provided by Diversity Arrays Technology, and the covfilename specifies
the csv file which contains metrics associated with each individual (id, pop, sex, etc).
Using the example data set provided in the package (accessed via an internal path to the files)

dartfile <- system.file("extdata","testset_SNPs_2Row.csv", package="dartR")

metafile <- system.file("extdata","testset_metadata.csv", package="dartR")

gl <- gl.read.dart(filename=dartfile, ind.metafile = metafile, probar=FALSE)

## Starting gl.read.dart
## Starting utils.read.dart
## Topskip not provided. Setting topskip to 3 .
## Reading in the SNP data
## Detected 2 row format.
## Added the following locus metrics:
## AlleleID CloneID AlleleSequence SNP SnpPosition CallRate OneRatioRef OneRatioSnp TrimmedSequence Freq
## Number of rows per clone (should be only 2 s): 2
## Recognised: 250 individuals and 255 SNPs in a 2 row format using C:/Users/s425824/AppData/Local/Tem
## Completed: utils.read.dart
## Starting utils.dart2genlight
## Starting conversion....
## Format is 2 rows.
## Please note conversion of bigger data sets will take some time!
## Once finished, we recommend to save the object using save(object, file="object.rdata")
## Adding individual metrics: C:/Users/s425824/AppData/Local/Temp/Rtmpk5JfRw/Rinst2b6c348a8f/dartR/extda
## Ids for individual metadata (at least a subset of) are matching!
## Found 250 matching ids out of 250 ids provided in the ind.metadata file.
## Added population assignments.
## Added latlon data
## Added id to the other$ind.metrics slot.
## Added pop to the other$ind.metrics slot.
## Added lat to the other$ind.metrics slot.
## Added lon to the other$ind.metrics slot.
## Added sex to the other$ind.metrics slot.
## Added maturity to the other$ind.metrics slot.
## Completed: utils.dart2genlight
## Data read in. Please check carefully the output above
## Read depth calculated and added to the locus metrics
## Minor Allele Frequency (maf) calculated and added to the locus metrics
## Completed: gl.read.dart

9
The resultant genlight object, gl, can be interrogated to determine if the data have been input correctly.
To dispay (parts of) the genlight object we have the following options:
Display the structure of the genlight object

gl

## /// GENLIGHT OBJECT /////////

##
## // 250 genotypes, 255 binary SNPs, size: 749.5 Kb
## 7868 (12.34 %) missing data
##
## // Basic content
## @gen: list of 250 SNPbin
## @ploidy: ploidy of each individual (range: 2-2)
##
## // Optional content
## @ind.names: 250 individual labels
## @loc.names: 255 locus labels
## @loc.all: 255 alleles
## @position: integer storing positions of the SNPs
## @pop: population of each individual (group size range: 1-11)
## @other: a list containing: loc.metrics latlong ind.metrics loc.metrics.flags verbose history

Display the SNP genotypes for the first 3 individuals and 5 loci

as.matrix(gl)[1:3,1:3]

## 100049687-12-C/T 100049698-16-G/A 100049728-23-A/G

## AA010915 2 NA 0
## UC_00126 2 NA 0
## AA032760 NA NA 0

Report the number of loci, individuals and populations

nLoc(gl)

## [1] 255

nInd(gl)

## [1] 250

nPop(gl)

## [1] 30

List the population labels (only the first 5)

10
levels(pop(gl))[1:5]

## [1] "EmmacBrisWive" "EmmacBurdMist" "EmmacBurnBara" "EmmacClarJack"

## [5] "EmmacClarYate"

To load your own data in, you need the csv file provided by DArT (e.g. mydata.csv) and (optional) an
individual metadata file (e.g. my.metadata.csv).

gl <- gl.read.dart(filename="mydata.csv", ind.metafile = "my.metadata.csv")

5 Working with DArT Genlight Objects

Genlight objects for working with DArT genotypes have all of the general attributes described above, but
also have some specific characteristics that derive from the DArT preparatory filtering and associated quality
statistics. This is shown diagrammatically below.

11
5.1 Locus metadata

The locus metadata included in the genlight object are those provided as part of your DArT report. These
metadata are obtained from the DArT 1Row or 2Row csv file when it is read in to the genlight object. The
locus metadata are held in an R data.frame that is associated with the SNP data as part of the genlight
object.
The locus metadata would typically include:

identifier explanation
CloneID: Unique identifier for the sequence in which the SNP marker occurs

12
 identifier explanation
SNP: In 2-row format, this column is blank in the Reference row, and contains the base
position and base variant details in the SNP row. In 1-row format, this column
contains the base position and base variant details
SnpPosition: The position (zero is position 1) in the sequence tag at which the defined SNP
variant base occurs
TrimmedSequence (optional) The sequence containing the SNP or SNPs, trimmed of adaptors.
CallRate: The proportion of samples for which the genotype call is non-missing (that is, not
“-” )
OneRatioRef: The proportion of samples for which the genotype score is “1”, in the Reference
allele row of the 2-row format.
OneRatioSnp: The proportion of samples for which the genotype score is “1”, in the SNP allele
row of the 2-row format
FreqHomRef: The proportion of samples homozygous for the Reference allele
FreqHomSnp: The proportion of samples homozygous for the Alternate (SNP) allele
FreqHets: The proportion of samples which score as heterozygous.
PICRef: The polymorphism information content (PIC) for the Reference allele row
PICSnp: The polymorphism information content (PIC) for the SNP allele row
AvgPIC: The average of the polymorphism information content (PIC) of the Reference and
SNP allele rows
AvgCountRef: The sum of the tag read counts for all samples, divided by the number of samples
with non-zero tag read counts, for the Reference allele row
AvgCountSnp: The sum of the tag read counts for all samples, divided by the number of samples
with non-zero tag read counts, for the Alternate (SNP) allele row
RepAvg: The proportion of technical replicate assay pairs for which the marker score is
consistent.

These metadata variables are held in the genlight object as an associated data.frame called loc.metrics, which
can be accessed in the following form:

#Only the entries for the first ten individuals are shown
gl@other$loc.metrics$RepAvg[1:10]

## [1] 1.000000 1.000000 1.000000 1.000000 0.989950 1.000000 0.993274 1.000000

## [9] 1.000000 0.980000

You can check the names of all available loc.metrics via:

names(gl@other$loc.metrics)

## [1] "AlleleID" "CloneID" "AlleleSequence" "SNP"

## [5] "SnpPosition" "CallRate" "OneRatioRef" "OneRatioSnp"
## [9] "TrimmedSequence" "FreqHomRef" "FreqHomSnp" "FreqHets"
## [13] "PICRef" "PICSnp" "AvgPIC" "AvgCountRef"
## [17] "AvgCountSnp" "RepAvg" "clone" "uid"
## [21] "rdepth"

Depending on the report from DarT you may have additional (fewer) loc.metrics (e.g. Trimmed Sequence is
available on request).
These metadata are used by the {dartR} package for various purposes, so if any are missing from your
dataset, then there will be some analyses that will not be possible. For example, TrimmedSequence is

13
used to generate output for subsequent phylogenetic analyses that require estimates of base frequencies and
transition and transversion ratios.
CloneID is essential (with its very special format), and dartR scripts for loading your data sets will terminate
with an error message if this is not present.

5.2 Individual Metadata

Individual (=specimen or sample) metadata are user specified, and do not come from DArT. Individual
metadata are held in a second dataframe associated with the SNP data in the genlight object. See the figure
above.
Two special individual metrics are:

individual
metric explanation
id Unique identifier for the individual or specimen that links back to the physical sample
pop A label for the biological population from which the individual was drawn

These metrics are supplied by the user by way of a metafile, provided at the time of inputting the SNP data
to the genlight object. A metafile is a comma-delimited file, usually named ind_metrics.csv or similar, that
contains labelled columns. The file must have a column headed id, which contains the individual (=specimen
or sample labels) and a column headed pop, which contains the populations to which individuals are assigned.
These special metrics can be accessed a:
gl@pop or pop(gl)
and
[email protected] or indNames(gl)
A number of other user-defined metrics can be included in the metadata file. Examples of user-defined
metadata for individuals include:

metric explanation
sex Sex of the individual (Male, Female)
maturity Maturity of the individual (Adult, Subadult, juvenile)
lat Latitude of the location of collection
long Longitude of the location of collection

These optional data are provided by the user in the same metafile used to assign id labels and assign
individuals to populations. It is the excel csv file referred to above.
The individual metadata are held in the genlight object as a dataframe named ind.metrics and can be
accessed using the following form:

#only first 10 entries showns

gl@other$ind.metrics$sex[1:10]

## [1] Male Male Male Male Unknown Male Female Female Male
## [10] Female
## Levels: Female Male Unknown

14
5.3 Importing data from other formats/sources

In the table below we list possible avenues to load data from other formats and most importantly also from
text simple files, which opens the package in principle to any other source.

Import path Package Pathway* Description

gl.read.dart dartR — based on DaRT data [with optional
meta data for individuals]
read.loci pegas loci2genind, data set are provided as a csv text
gi2gl file (?read.loci)
read.vcfR pegas vcfR2genlight vcf text file (vcfR package)
read.fstat adegenet gi2gl Fstat format (version 2.9.3) by
Jerome Goudet
read.genetix adegenet gi2gl Format Belkhir K., Borsa P., Chikhi
L., Raufaste N. & Bonhomme F.
(1996-2004) GENETIX
read.structure adegenet gi2gl Structure format of Pritchard, J.;
Stephens, M. & Donnelly, P. (2000)
read.PLINK adegenet — Data provided in PLINK format
fasta2genlight adegenet — Extracts SNPs data from fasta
format (?adegenet)
read.genetable PopGenReport gi2gl csv text file based on df2genind
Adamack & Gruber (2014)
(?read.genetable)

*Pathway provides the order of functions needed to convert data to genlight, — indicates that the function
direclty converts to a genlight object
Below we exemplify an import from a text file. Provided with the package is a simple data set in a text file
format (editable via Excel or Calc) or any text editor. To convert your data into a genlight object you can
arrange your data set in the same format and also attach meta data for loci such as call rates, sequences and
individuals (will be stored in the @other slot under), such as coordinates, lat/lons and population structure.
To have a look at the format in the provided file, type:

read.csv( paste(.libPaths()[1],"/dartR/extdata/platy.csv",sep="" ))

## ind pop lat long group age loci1 loci2 loci3 loci4 loci5
## 1 T158 Black -40.86642 145.2836 Female juv A/A G/C A/T A/A G/C
## 2 T306 Black -40.85589 145.2764 Male Ad A/A G/G A/A A/A G/C
## 3 T305 Black -40.87889 145.2885 Female Ad A/A G/G A/T A/A G/C
## 4 T148 Black -40.99193 145.3757 Male Ad A/A G/G A/A A/A G/C
## 5 T149 Black -40.99193 145.3757 Female Ad A/A G/C A/T A/A G/C
## 6 T106 Brid -41.23205 147.4597 Male Ad A/A G/G A/A A/A G/G
## 7 T107 Brid -41.23205 147.4597 Male Ad A/A G/G A/T A/A G/G
## 8 T110 Brid -41.23205 147.4597 Female Ad A/A G/G A/A A/A G/G
## 9 T111 Brid -41.23205 147.4597 Female Ad A/T G/C A/T A/A G/G
## 10 T308 Cam -41.09567 145.7958 Male Sub-Ad A/T C/C A/A A/A G/G
## 11 T307 Cam -41.06975 145.8152 Male Ad A/T C/C A/T A/A C/C
## 12 T302 Cam -41.05121 145.8280 Male Ad T/T G/C A/A A/A C/C
## 13 T303 Cam -41.04764 145.8230 Female Juv T/T C/C A/T A/A C/C
## loci6
## 1 T/A

15
## 2 T/A
## 3 T/A
## 4 T/A
## 5 T/A
## 6 T/A
## 7 T/A
## 8 T/A
## 9 T/A
## 10 T/A
## 11 T/A
## 12 T/T
## 13 A/A

We now have to specify each of the columns to be able to read it into R as a genind format using
read.genetable from the PopGenReport package. For a detailed explanation see ?read.genetable.

#you might need to install PopGenReport via

#install.packages("PopGenReport")
library(PopGenReport)

## Loading required package: knitr

platy <- read.genetable( paste(.libPaths()[1],"/dartR/extdata/platy.csv",

sep="" ), ind=1, pop=2, lat=3, long=4, other.min=5, other.max=6, oneColPerAll=FALSE,sep="/")

platy

## /// GENIND OBJECT /////////

##
## // 13 individuals; 6 loci; 11 alleles; size: 13.1 Kb
##
## // Basic content
## @tab: 13 x 11 matrix of allele counts
## @loc.n.all: number of alleles per locus (range: 1-2)
## @loc.fac: locus factor for the 11 columns of @tab
## @all.names: list of allele names for each locus
## @ploidy: ploidy of each individual (range: 2-2)
## @type: codom
## @call: df2genind(X = genes, sep = sep, ncode = ncode, ind.names = as.character(inds),
## loc.names = colnames(genes), pop = as.character(pops), NA.char = NA.char,
## ploidy = ploidy)
##
## // Optional content
## @pop: population of each individual (group size range: 4-5)
## @other: a list containing: latlong data

To convert the genind to a genlight object simply type:

platy.gl <- (gi2gl(platy))

## Starting gi2gl
## Completed: gi2gl

16
The resulting genlight object has already some of the meta data as provided in the table but not in the right
place (pop, indNames ploidy, loc.n.all and latlong in the “@other“‘ slot are already there).
For example to put the individual meta data (group, age) into the right slot in a genlight object we need to
move it into the @other$ind.metrcis slot.

platy.gl@other$ind.metrics <- platy.gl@other$data

If there are additional data available for the loci (e.g. via a csv file) we can also attach that via the
@other$loc.metrics slot using read.csv. For demonstratation we will create some random loci meta
data with the code below and store it into a data.frame called df.loc. We create two columns: Trimmed-
Sequnce (containing the sequence data for each loci, trimmed to a consistent length, and removing the
adaptors), and a quality index (ranging from zero to one indicating the reproducibility of repeated runs for
that loci).

ts <- sapply(1:nLoc(platy.gl), function(x) paste(sample(c("A","T","G","C"), 50, replace = T),

collapse = ""))
df.loc <- data.frame(RepAvg = runif(nLoc(platy.gl)), TrimmedSequence=ts)

platy.gl@other$loc.metrics <- df.loc

The last line of code stores the loci metadata in the right location and we can use our filter and report
function now on the genlight object as before.

gl.report.callrate(platy.gl)

## Starting gl.report.callrate
## Processing a SNP dataset
## Warning: genlight object contains monomorphic loci which will be factored into Callrate calculation

17
 SNP data − Call Rate by Locus
1.0

0.5

0.0

−0.5

−1.0
0.00 0.25 0.50 0.75 1.00

4
count

0
0.00 0.25 0.50 0.75 1.00
callrate

## Reporting Call Rate by Locus

## No. of loci = 6
## No. of individuals = 13
## Miniumum Call Rate: 1
## Maximum Call Rate: 1
## Average Call Rate: 1
## Missing Rate Overall: 0
## Completed: gl.report.callrate

## $callrates
## Threshold Retained Percent Filtered Percent
## 1 1.00 6 100 0 0
## 2 0.95 6 100 0 0
## 3 0.90 6 100 0 0
## 4 0.85 6 100 0 0
## 5 0.80 6 100 0 0
## 6 0.75 6 100 0 0
## 7 0.70 6 100 0 0
## 8 0.65 6 100 0 0
## 9 0.60 6 100 0 0
## 10 0.55 6 100 0 0
## 11 0.50 6 100 0 0
## 12 0.45 6 100 0 0
## 13 0.40 6 100 0 0
## 14 0.35 6 100 0 0

18
## 15 0.30 6 100 0 0
## 16 0.25 6 100 0 0
## 17 0.20 6 100 0 0
## 18 0.15 6 100 0 0
## 19 0.10 6 100 0 0
## 20 0.05 6 100 0 0
## 21 0.00 6 100 0 0
##
## $boxplot

SNP data − Call Rate by Locus

1.0

0.5

0.0

−0.5

−1.0

0.00 0.25 0.50 0.75 1.00

##
## $hist

19
 6

4
count

0.00 0.25 0.50 0.75 1.00

callrate

gl2 <- gl.filter.reproducibility(platy.gl, t=0.5)

## Starting gl.filter.reproducibility
## Processing a SNP dataset
## Identifying loci with repeatability below : 0.5
## Removing loci with repeatability less than 0.5
## Completed: gl.filter.reproducibility

Some of the functions require additional meta data, e.g. the gl2fasta need sequence data (TrimmedSequnce),
the position of the SNP in the sequence for each loci @position and the type of allelels for each loci @loc.all.
If such data is available the information needs to be stored in the correct slots to work with the function.
The code below exports our genlight object to a fasta format (again we create random meta data via the
code below).

platy.gl@position <- as.integer(runif(nLoc(platy.gl),2,49))

[email protected] <- [email protected][1:6]

And finally we can use the gl2fasta function (using method 1, see explanation in ?gl2fasta):

gl2fasta(platy.gl)

20
6 Subsetting and Recoding Data

6.1 Filtering

A range of filters are available for selecting individuals or loci on the basis of quality metrics.

function explanation
gl.report.callrate() Calculate and report the number of loci or individuals for which the call
rate exceeds a range of thresholds.
gl.filter.callrate() Calculate call rate (proportion with non-missing scores) for each locus or
individual and remove those loci or individuals below a specified threshold.
gl.report.repavg() Report the number of loci or individuals for which the reproducibility
(averaged over the two allelic states) exceeds a range of thresholds.
gl.filter.repavg() Remove those loci or individuals for which the reproducibility (averaged
over the two allelic states) falls below a specified threshold.
gl.report.secondaries() Report the number of sequence tags with multiple SNP loci, and the number
of SNP loci that are part of or individuals for which the reproducibility
(averaged over the two allelic states) exceeds a range of thresholds.
gl.filter.secondaries() Remove all but one locus where there is more than one locus per sequence
tag.
gl.report.monomorphs() Report the number of monomorphic loci and the number of loci for which
the scores are all missing (NA).
gl.filter.monomorphs() Remove all monomorphic loci, including loci for which the scores are all
missing (NA).
gl.report.hamming() Report the distribution of pairwise Hamming distances between trimmed
sequence tags.
gl.filter.hamming() Filter loci by taking out one of a pair of loci with Hamming distances less
than a threshold.
gl.filter.hwe Filters departure of Hardy-Weinberg-Equilibrium for every loci per
population or overall
gl.report.hwe Reports departure of Hardy-Weinberg-Equilibrium for every loci per
population or overall

Refer to the help on each function for details of the parameters taken by each of these functions using
?nameoffunction.

6.2 Examples of dartR code to filter a gl dataset

Filtering of data is often necessary to make sure only high quality loci (few missing data) and with a
consistent quality are retained and “noise” in the data set is minimised. In addition by filtering you reduce
the number of loci, which often speeds up the analysis considerably. The kind and order of filtering that
someone applies depends very much on the intended analysis. For example for classical calculation of indices
of population structure, loci and individuals with lots of missing data might be discarded, though for other
kind of analysis the amount of missing data may hint to “hidden” subspecies in the populations. Therefore
a general adivse on the order and amount of filtering cannot be given. As an example if the focus is towards
studying population structure, where only a limited number of individuals are sampled, a valid strategy is
to filter in such a way that the number of individuals per population are maintained, but the number of loci
can be reduced. So a suggested order here is:

1. Filter by reproducibility (gl.filter.reproducibility in dartR) (a meassurement of quality per loci)

21
 2. Fitler by monomorphic loci (gl.filter.monomorphs)(as they do not provide information for popula-
tion structure and simply slow the analysis)
3. Filter by amount of missing data (gl.filter.callrate, method="loc") per locus
4. Filter to remove all but one of multiple snps in the same fragment (gl.filter.secondaries)
5. Filter individuals by amount of missing data (gl.filter.callrate, method="ind")

Additional filters to apply could be for excluding possible loci under selection (gl.outflank), checking
loci for linkage disequilibrium (gl.report.ld) or filtering for loci out of Hardy-Weinberg-Equilibrium
(gl.filter.hwe)
Simple examples how to apply some of the filters are provided below.

1. Filter on call rate, threshold = at least 95% loci called

gl2 <- gl.filter.callrate(gl, method = "loc", threshold = 0.95)

## Starting gl.filter.callrate
## Processing a SNP dataset
## Warning: Dataset contains monomorphic loci which will be included in the Call Rate calculations for
## Removing loci based on Call Rate, threshold = 0.95

Call Rate by Individual

[pre−filtering, SNP dataset]
Frequency

0 100

0.4 0.5 0.6 0.7 0.8 0.9 1.0

Call Rate

[post−filtering]
Frequency

0 100

0.6 0.7 0.8 0.9 1.0

Call Rate

## Completed: gl.filter.callrate

2. Filter individuals on call rate (threshold =90% )

22
gl2 <- gl.filter.callrate(gl, method="ind", threshold = 0.90)

## Starting gl.filter.callrate
## Processing a SNP dataset
## Warning: Dataset contains monomorphic loci which will be included in the Call Rate calculations for
## Removing individuals based on Call Rate, threshold = 0.9
## No. of individuals deleted (CallRate <= 0.9 :
## AA010915[EmmacMDBForb], AA032760[EmmacMDBMaci], AA011723[EmmacBurnBara], AA012411[EmmacCoopEulb], AA0

Call Rate by Individual

[pre−filtering, SNP dataset]
Frequency

0 40

0.75 0.80 0.85 0.90 0.95 1.00

Call Rate

[post−filtering]
Frequency

8
0

0.90 0.92 0.94 0.96 0.98 1.00

Call Rate

## Note: Locus metrics not recalculated

## Note: Resultant monomorphic loci not deleted
## Completed: gl.filter.callrate

3. Filter on reproducibility, threshold (here called t, do not ask why) 100% reproducible

gl2 <- gl.filter.reproducibility(gl, t=1)

## Starting gl.filter.reproducibility
## Processing a SNP dataset
## Identifying loci with repeatability below : 1
## Removing loci with repeatability less than 1
## Completed: gl.filter.reproducibility

4. Filter out multiple snps in single sequence tags (!!!!!produces an error currently!!!!!)

23
 5. Filter out monomorphic loci

gl2 <- gl.filter.monomorphs(gl, v=0)

6. Filter out loci with trimmed sequence tags that are too similar (possible paralogues). Only works if
TrimmedSequence is available in the loci metadata, therefore we use another test data set here.

gl2 <- gl.filter.hamming(testset.gl, t=0.25, pb = F)

## Starting gl.filter.hamming
## Processing a SNP dataset
## Calculating Hamming distances between sequence tags
## Filtering loci with Hamming Distance is less than 0.25
## Completed: gl.filter.hamming

Note: This filter and its accompanying report function is slow when there are many loci. Recommended
that it be applied after all other filtering, and only if less than 20,000 loci remain. May require an overnight
run.
Please note in the examples we always stored the resulting filter into a new genlight object gl2, gl3 etc.
Though it is a bit of a waste in terms of memory, it avoids confusion which filter you have already applied.
A series of filter could then look like:

gl2 <- gl.filter.callrate(gl, method = "loc", threshold = 0.95)

gl3 <- gl.filter.callrate(gl2, method="ind", threshold = 0.90)
gl4 <- gl.filter.repavg(gl3, t=1)

Note that filters that result in the removal of populations or individual have optional parameters to request
that the locus metadata be recalculated or for resultant monomorphic loci to be removed. Recalculation
of the locus metadata is necessary because callrate, for example, will no longer be accurate once some
individuals have been removed from the dataset.

7 Subsetting and Recoding Data

7.1 Population (=higher level grouping) reassignment

Recall that the metadata file provided when the data are initially input contains information assigned to
each individual including, often at a minimum, population assignment. There are various ways to reassign
individuals to populations, rename populations or individuals, delete populations or individuals after the
data have been read in to a genlight object.
The initial population assignments via the metafile can be viewed via:

#population names (#30 populations)

levels(pop(gl))

## [1] "EmmacBrisWive" "EmmacBurdMist" "EmmacBurnBara" "EmmacClarJack"

## [5] "EmmacClarYate" "EmmacCoopAvin" "EmmacCoopCully" "EmmacCoopEulb"
## [9] "EmmacFitzAllig" "EmmacJohnWari" "EmmacMDBBowm" "EmmacMDBCond"
## [13] "EmmacMDBCudg" "EmmacMDBForb" "EmmacMDBGwyd" "EmmacMDBMaci"

24
## [17] "EmmacMDBMurrMung" "EmmacMDBSanf" "EmmacMaclGeor" "EmmacMaryBoru"
## [21] "EmmacMaryPetr" "EmmacNormJack" "EmmacNormLeic" "EmmacNormSalt"
## [25] "EmmacRichCasi" "EmmacRoss" "EmmacRussEube" "EmmacTweeUki"
## [29] "EmsubRopeMata" "EmvicVictJasp"

#table on individuals per population

table(pop(gl))

##
## EmmacBrisWive EmmacBurdMist EmmacBurnBara EmmacClarJack
## 10 10 11 5
## EmmacClarYate EmmacCoopAvin EmmacCoopCully EmmacCoopEulb
## 5 10 10 10
## EmmacFitzAllig EmmacJohnWari EmmacMDBBowm EmmacMDBCond
## 10 10 10 10
## EmmacMDBCudg EmmacMDBForb EmmacMDBGwyd EmmacMDBMaci
## 10 11 9 10
## EmmacMDBMurrMung EmmacMDBSanf EmmacMaclGeor EmmacMaryBoru
## 10 10 11 6
## EmmacMaryPetr EmmacNormJack EmmacNormLeic EmmacNormSalt
## 4 6 1 1
## EmmacRichCasi EmmacRoss EmmacRussEube EmmacTweeUki
## 10 10 10 10
## EmsubRopeMata EmvicVictJasp
## 5 5

It is easy to create a barplot on the number of individuals per population:

barplot(table(pop(gl)), las=2)

25
 10

0
EmmacBrisWive
EmmacBurdMist
EmmacBurnBara
EmmacClarJack
EmmacClarYate
EmmacCoopAvin
EmmacCoopCully
EmmacCoopEulb
EmmacFitzAllig
EmmacJohnWari
EmmacMDBBowm
EmmacMDBCond
EmmacMDBCudg
EmmacMDBForb
EmmacMDBGwyd
EmmacMDBMaci
EmmacMDBMurrMung
EmmacMDBSanf
EmmacMaclGeor
EmmacMaryBoru
EmmacMaryPetr
EmmacNormJack
EmmacNormLeic
EmmacNormSalt
EmmacRichCasi
EmmacRoss
EmmacRussEube
EmmacTweeUki
EmsubRopeMata
EmvicVictJasp
If you have only a few reassignments, the simplest way is to use one or more of the scripts
gl <- gl.keep.pop(gl, c(pop1, pop5, pop7)) # Retains only populations 1, 5 and 7 gl <- gl.drop.pop(gl,
c(pop2, pop3, pop4, pop6)) # Deletes populations 2, 3, 4, 6 gl <- gl.merge.pop(gl, old=c(“pop1”,“pop2”),
new=“pop1”) # Merges populations 1 and 2 as pop1 gl <- gl.merge.pop(gl, old=c(“pop1”), new=“outgroup”)
# Renames population 1 to a population labelled outgroup
Try these out for yourself.
If only a few populations are involved, then the best option is to use the gl.drop.pop or gl.keep.pop functions.

glnew3 <- gl.keep.pop(gl, pop.list=c("EmsubRopeMata","EmvicVictJasp"))

will delete all individuals in all populations except those listed.

glnew3 <- gl.drop.pop(gl, pop.list=c("EmsubRopeMata","EmvicVictJasp"))

will delete all individuals in the listed populations.

glnew3 <- gl.merge.pop(gl, old=c("EmsubRopeMata","EmvicVictJasp"), new="outgroup")

will reassign individuals in populations EmsubRopeMata and EmvicVictJasp to a new population called
outgroup.

26
glnew3 <- gl.merge.pop(gl, old="EmsubRopeMata", new="Emydura_victoriae")

will rename population EmvicVictJasp to Emydura_victoriae.

Note that there is an option for recalculating the relevant individual metadata, and for removing resultant
monomorphic loci.

7.2 Individual reassignment

You can reassign individuals to new populations in a number of ways.A similar set of scripts apply to
individuals.
The initial individual labels entered at the time of reading the data into the genlight object can be viewed
via:

#individual names
indNames(gl)

## [1] "AA010915" "UC_00126" "AA032760" "AA013214" "AA011723" "AA012411"

## [7] "AA019237" "AA019238" "AA019239" "AA019235" "AA019240" "AA019241"
## [13] "AA019242" "AA019243" "AA019251" "AA019252" "AA012405" "AA012406"
## [19] "AA012409" "AA012499" "AA012422" "AA012434" "AA012469" "AA012500"
## [25] "AA032799" "AA032826" "AA010795" "AA010796" "AA032800" "AA032801"
## [31] "AA032808" "AA032809" "AA032811" "AA032812" "AA032822" "AA032825"
## [37] "AA010797" "AA010752" "AA010754" "AA010756" "AA010798" "AA010799"
## [43] "AA010800" "AA010802" "AA010803" "AA010804" "AA010809" "AA010749"
## [49] "AA010758" "AA010763" "AA010765" "AA010771" "AA010772" "AA010781"
## [55] "AA032762" "AA032763" "AA032756" "AA032757" "AA032758" "AA032761"
## [61] "AA032765" "AA010931" "AA010937" "AA010940" "AA032764" "AA032768"
## [67] "AA010936" "AA010909" "AA010916" "AA010917" "AA010920" "AA010921"
## [73] "AA020651" "AA020652" "AA020667" "AA020669" "AA020655" "AA020656"
## [79] "AA020644" "AA020645" "AA020646" "AA020649" "AA013203" "AA013217"
## [85] "AA013220" "AA013202" "AA013225" "AA018496" "AA018497" "AA018513"
## [91] "AA013231" "AA013261" "AA013265" "AA013270" "AA018492" "AA018493"
## [97] "AA018494" "AA018495" "AA018514" "AA018515" "AA018516" "UC_00125"
## [103] "UC_00126a" "UC_00146" "UC_00149" "AA018640" "AA018658" "AA011729"
## [109] "UC_00132" "UC_00137" "UC_00143" "UC_00157" "UC_00161" "AA018637"
## [115] "AA018638" "AA018639" "AA011731" "AA033576" "AA033577" "AA011732"
## [121] "AA011737" "AA011741" "AA011744" "AA011745" "AA011746" "AA011749"
## [127] "AA033575" "AA033578" "AA012411a" "AA033579" "AA033582" "AA033593"
## [133] "AA033602" "AA033609" "AA033617" "AA010915a" "AA011723a" "AA019158"
## [139] "AA020379" "UC_01044" "AA018380" "AA018371" "AA004553" "AA000328"
## [145] "AA000311" "AA019159" "AA020378" "UC_01060" "AA018379" "AA018365"
## [151] "AA004554" "AA000303" "AA000320" "AA019160" "AA020377" "UC_01053"
## [157] "AA018375" "AA004555" "AA000304" "AA019165" "AA019161" "AA020376"
## [163] "UC_01062" "AA018374" "AA04523" "AA000305" "AA019164" "AA020375"
## [169] "AA018373" "AA032875" "AA000309" "AA019163" "AA020374" "AA018368"
## [175] "AA032878" "AA000302" "AA019162" "AA020365" "UC_00150" "AA018369"
## [181] "AA004551" "AA032880" "AA000307" "AA019156" "AA020371" "UC_01051"
## [187] "AA018370" "AA004552" "AA032882" "AA000310" "AA019157" "AA019075"
## [193] "AA004864" "AA019071" "AA004868" "AA019083" "AA019072" "AA004858"
## [199] "AA004869" "AA019082" "AA019073" "AA004859" "AA004866" "AA019077"
## [205] "AA004860" "AA019080" "AA004861" "AA019079" "AA004862" "AA019078"

27
## [211] "AA004863" "UC_00267" "UC_00205" "UC_00206" "UC_00208" "UC_00243"
## [217] "UC_00209" "UC_00254" "UC_00210" "UC_00259" "UC_00126c" "AA063718"
## [223] "AA063720" "AA063722" "AA063726" "AA063732" "AA063708" "AA063710"
## [229] "AA063712" "AA063714" "AA063716" "AA020735" "AA032442" "AA032441"
## [235] "AA020749" "AA020746" "AA020744" "AA020743" "AA020739" "AA020738"
## [241] "AA001451" "AA01452" "AA001454" "AA001455" "AA001446" "AA001456"
## [247] "AA001447" "AA001448" "AA001449" "AA001450"

The individuals can be manipulated using

gl <- gl.keep.ind(gl, c(ind1, ind5, ind7)) # Retains only individuals labelled ind1, 5 and 7 gl <- gl.drop.pop(gl,
c(ind2, ind3, ind4, ind6)) # Deletes populations 2, 3, 4, 6
Try these out for yourself, by listing the individuals using indNames() and then deleting a selected few.

glnew3 <- gl.keep.ind(gl, ind.list=c("AA019073","AA004859"))

will delete all individuals except those listed.

glnew3 <- gl.drop.pop(gl, ind.list=c("AA019073","AA004859"))

will delete all individuals listed.

7.3 Recode tables

Alternatively, reassignment and deletion of populations can be effected using a recode table, that is, a table
stored as a csv file containing the old population assignments and the new population assignments as two
columns. The quickest way to construct a recode table for an active genlight object is using

gl.make.recode.pop(gl, outfile = "new_pop_assignments.csv")

Hint

Please note we are using the tempdir() to read/write files to a location in all examples. Feel free to
change that to your needs by just providing a path to the folder of your liking. Normally, this would
be your working directory specified with setwd().

This will generate a csv file with two columns, the first containing the existing population assignments, and
the second also containing those assignments ready for editing to achieve the reassignments. This editing is
best done in Excel.
The population reassignments are then applied using:

glnew <- gl.recode.pop(gl, pop.recode="new_pop_assignments.csv")

You can check that the new assignments have been applied with:

28
levels(pop(gl))

## [1] "EmmacBrisWive" "EmmacBurdMist" "EmmacBurnBara" "EmmacClarJack"

## [5] "EmmacClarYate" "EmmacCoopAvin" "EmmacCoopCully" "EmmacCoopEulb"
## [9] "EmmacFitzAllig" "EmmacJohnWari" "EmmacMDBBowm" "EmmacMDBCond"
## [13] "EmmacMDBCudg" "EmmacMDBForb" "EmmacMDBGwyd" "EmmacMDBMaci"
## [17] "EmmacMDBMurrMung" "EmmacMDBSanf" "EmmacMaclGeor" "EmmacMaryBoru"
## [21] "EmmacMaryPetr" "EmmacNormJack" "EmmacNormLeic" "EmmacNormSalt"
## [25] "EmmacRichCasi" "EmmacRoss" "EmmacRussEube" "EmmacTweeUki"
## [29] "EmsubRopeMata" "EmvicVictJasp"

Task
Try this using commands in the R editor to create the comma-delimited recode file, edit in in Excel
to remove the Emmac prefix from populations, then apply it using the above command from the R
editor. Check your results.

Another way of population reassignment is to use:

glnew2 <- gl.edit.recode.pop(gl)

This command will bring up a window with a table showing the existing population assignments, with a
second column available for editing. When the window is closed, the assignments will be applied. If you
have optionally nominated a pop.recode file, a recode table will be written to file for future use.
Again, you can check that the new assignments have been applied with levels(pop(gl)).
###Deleting populations with a recode table
You can delete selected populations from a genlight object using the “Delete” keyword in the population
recode file. By reassigning populations to Delete, you are flagging them for deletion, and when the recode
table is applied, individuals belonging to those populations will be deleted from the genlight object, and any
resultant monomorphic loci will be removed.
Again, you can check that the new assignments have been applied and requested populations deleted with
levels(pop(gl)).

Task

Try deleting some populations, say the outgroup populations (EmsubRopeMata and EmvicVictJasp)
using gl.edit.recode.pop()from the R editor. Check your results for example using:
table(pop(gl))

7.3.1 Relabeling and deleting individuals with a recode table

Recall that the genlight object contains labels for each individual. It obtains these names from the csv
datafile provided by DArT at the time of reading these data in. There may be reasons for changing these

29
individual labels - there may have been a mistake, or new names need to be provided in preparation for
analyses to be included in publications.
Individual recode tables are csv files (comma delimited text files) that can be used to rename individuals in
the genlight object or for deleting individuals from the genlight object. These population assignments can
be viewed using
#only first 10 entries are shown
indNames(gl)[1:10]

## [1] "AA010915" "UC_00126" "AA032760" "AA013214" "AA011723" "AA012411"

## [7] "AA019237" "AA019238" "AA019239" "AA019235"

The quickest way to rename individuals is to construct a recode table for an active genlight object is using

gl.make.recode.ind(gl, outfile="new_ind_assignments.csv")

This will generate a csv file with two columns, the first containing the existing individual names, and the
second also containing those names ready for editing. This editing is best done in Excel.
The population reassignments are then applied using

glnew3 <- gl.recode.ind(gl, ind.recode="new_ind_assignments.csv")

You can check that the new assignments have been applied with indNames(gl)
Another way of individual reassignment is to use

gl <- gl.edit.recode.ind(gl, ind.recode="new_ind_assignments.csv")

This command will bring up a window with a table showing the existing individual labels, with a second
column available for editing. When the window is closed, the renaming will be applied. If you have optionally
nominated a ind.recode file, a recode table will be written to file for future use. Again, you can check that
the new assignments have been applied with indNames(gl).

7.4 Deleting individuals

You can delete selected individuals from a genlight object using the “Delete” keyword in the individual
recode file. By renaming individuals to Delete, you are flagging them for deletion, and when the recode table
is applied, those individuals will be deleted from the genlight object, and any resultant monomorphic loci
will be removed.
Again, you can check that the new assignments have been applied and requested populations deleted with
indNames(gl).
Note that there are options for recalculating the relevant individual metadata, and for removing resultant
monomorphic loci.

7.5 Recalculating locus metadata

When you delete individuals or populations, many of the locus metadata (such as Call Rate) are no longer
correct. You can recalculate the locus metadata using the script
gl <- gl.recalc.metrics(gl)
This is obviously important if you are drawing upon locus metadata in your calculations or filtering. The
script will also optionally remove monomorphic loci.

30
7.6 Using R commands to manipulate the genlight object

With your data in a genlight object, you have the full capabilities of the adegenet package at your fingertips
for subsetting your data, deleting SNP loci and individuals, selecting and deleting populations, and for
recoding to amalgamate or split populations. Refer to the manual [Analysing genome-wide SNP data using
adegenet] (http: //adegenet.r-forge.r-project.org/files/tutorial-genomics.pdf). For example:

gl_new <- gl[gl$pop!="EmmacBrisWive", ]

removes all individuals of the population EmmacBrisWive from the data set.
Note that this manual approach will not recalculate the individual metadata nor will it remove resultant
monomorphic loci. There are also some challenges with keeping the individual metadata matching the
individual records (see below).
The basic idea is here that we can use the indexing function [ ] on the genlight object gl to subset our
data set by individuals(=rows) and loci(=columns) in the same manner as we can subset a matrix in R.
For example:

glsub <- gl[1:7, 1:3]

glsub

## /// GENLIGHT OBJECT /////////

##
## // 7 genotypes, 3 binary SNPs, size: 240.4 Kb
## 8 (38.1 %) missing data
##
## // Basic content
## @gen: list of 7 SNPbin
## @ploidy: ploidy of each individual (range: 2-2)
##
## // Optional content
## @ind.names: 7 individual labels
## @loc.names: 3 locus labels
## @loc.all: 3 alleles
## @position: integer storing positions of the SNPs
## @pop: population of each individual (group size range: 1-1)
## @other: a list containing: loc.metrics latlong ind.metrics loc.metrics.flags verbose history

Subsets the data to the first seven individuals and the first three loci.
!!!Be aware that the accompanying meta data for individuals are subsetted, but the metadata for loci are
not!!!!. So if you check the dimensions of the meta data of the subsetted data set via:

dim(glsub@other$ind.metrics)

## [1] 7 6

dim(glsub@other$loc.metrics)

## [1] 255 21

31
you see that the subsetting of the meta data for individuals worked fine (we have seven indivduals (=rows)).
But we have still all the metadata for all loci (in the rows for the (=107 instead of 3). This “bug/feature”
is how the adegenet package implemented the genlight object.
To take care for the correct filtering for loci and individuals we suggest therefore to use the following approach:

1. create an index for individuals (if you want to subset by individuals)

2. create an index for loci (if you want to subset by loci))

For example you want to have only individuals of two populations (“EmmacRussEube” or “EmvicVictJasp”)
and 30 randomly selected loci you could type:

index.ind <- pop(gl)=="EmmacRussEube" | pop(gl)=="EmvicVictJasp"

#check if the index worked
table( pop(gl), index.ind)

## index.ind
## FALSE TRUE
## EmmacBrisWive 10 0
## EmmacBurdMist 10 0
## EmmacBurnBara 11 0
## EmmacClarJack 5 0
## EmmacClarYate 5 0
## EmmacCoopAvin 10 0
## EmmacCoopCully 10 0
## EmmacCoopEulb 10 0
## EmmacFitzAllig 10 0
## EmmacJohnWari 10 0
## EmmacMDBBowm 10 0
## EmmacMDBCond 10 0
## EmmacMDBCudg 10 0
## EmmacMDBForb 11 0
## EmmacMDBGwyd 9 0
## EmmacMDBMaci 10 0
## EmmacMDBMurrMung 10 0
## EmmacMDBSanf 10 0
## EmmacMaclGeor 11 0
## EmmacMaryBoru 6 0
## EmmacMaryPetr 4 0
## EmmacNormJack 6 0
## EmmacNormLeic 1 0
## EmmacNormSalt 1 0
## EmmacRichCasi 10 0
## EmmacRoss 10 0
## EmmacRussEube 0 10
## EmmacTweeUki 10 0
## EmsubRopeMata 5 0
## EmvicVictJasp 0 5

index.loc <- sample(nLoc(gl), 30, replace = F)

index.loc

## [1] 255 11 103 173 62 14 110 144 58 80 127 107 248 73 104 236 116 170 92
## [20] 254 132 185 25 85 141 133 243 6 56 70

32
and then

3. apply the indices to the genlight object and the meta data at the same time:

glsub2 <- gl[index.ind, index.loc]

glsub2@other$ind.metrics <- gl@other$ind.metrics[index.ind,] #not necessary
glsub2@other$loc.metrics <- gl@other$loc.metrics[index.loc,] #necessary

We can check the result via:

glsub2

## /// GENLIGHT OBJECT /////////

##
## // 15 genotypes, 30 binary SNPs, size: 197.7 Kb
## 53 (11.78 %) missing data
##
## // Basic content
## @gen: list of 15 SNPbin
## @ploidy: ploidy of each individual (range: 2-2)
##
## // Optional content
## @ind.names: 15 individual labels
## @loc.names: 30 locus labels
## @loc.all: 30 alleles
## @position: integer storing positions of the SNPs
## @pop: population of each individual (group size range: 5-10)
## @other: a list containing: loc.metrics latlong ind.metrics loc.metrics.flags verbose history

dim(glsub2@other$ind.metrics)

## [1] 15 6

dim(glsub2@other$loc.metrics)

## [1] 30 21

For those not fully versed in R, there are the above {dartR} filters to achieve the same end and the advantage
is that the filters do handle subsets of data correctly without any additional need to subset the meta data.
The advantage of the R approach is that it is much more useful in case you want to script your analysis
without intervention of a user when recoding your data set.

8 Genetic Distance
SNP data are multivariable data, in the sense that each individual (entity) has an allele profile (state) for
each of several loci (attributes). It is a simple data matrix because SNPs are bi-allelic, that is, each locus
can have one of three allelic states - aa, ab or bb, coded in a genlight object as 0, 1 or 2 respectively. One
allele (the most common allele) is assigned to the reference allele, and the other to the alternate (or in DArT
documentation, the SNP allele).

33
An obvious first choice in exploring the data is to construct a distance matrix between individuals based on
the genetic profiles, or to construct a distance matrix between populations based on their allele frequency
profiles.
There are a very many measures of genetic distance, but they collapse in number when applied to bi-allelic
SNP data. For example, Rogers D and Euclidean D differ only by a constant multiplier when the data are
biallelic. Distance matrices can be generated by a number of R packages, the most popular of which are
dist() from package stats and vegdist from package vegan. The function gl.dist.pop is a wrapper for
those two functions, applying them to allele frequencies calculated for each locus at each population defined
in the genlight object.
To calculate Euclidean distances (only for the first seven individuals and the first 100 loci, due to runtime
reasons):

gg <- gl[1:7,1:100]
#need to subset loc.metrics as well!!!
gg@other$loc.metrics <- gg@other$loc.metrics[1:100,]

gl.dist.pop(gg, method="euclidean")

## Starting gl.dist.pop
## Processing a SNP dataset
## Calculating distances: euclidean
## Refer to dist {stats} documentation for algorithm

## Standard boxplot, no adjustment for skewness

SNP data (DArTSeq)

Population euclidean Distances

0.0 0.1 0.2 0.3 0.4 0.5

15
10
5
0

0.0 0.1 0.2 0.3 0.4 0.5

34
##
## Reporting inter-population distances
## Distance measure: euclidean
## No. of populations = 7
## Average no. of individuals per population = 1
## No. of loci = 100
## Miniumum Distance: 0
## Maximum Distance: 0.54
## Average Distance: 0.127
##
## Completed: gl.dist.pop

## EmmacBurdMist EmmacBurnBara EmmacCoopEulb EmmacMDBForb

## EmmacBurnBara 0.0000000
## EmmacCoopEulb 0.0000000 0.0000000
## EmmacMDBForb 0.0000000 0.0000000 0.0000000
## EmmacMDBMaci 0.0000000 0.0000000 0.0000000 0.0000000
## EmmacMDBSanf 0.0000000 0.0000000 0.0000000 0.0000000
## EmmacMaclGeor 0.5330018 0.5360563 0.5330018 0.5391639
## EmmacMDBMaci EmmacMDBSanf
## EmmacBurnBara
## EmmacCoopEulb
## EmmacMDBForb
## EmmacMDBMaci
## EmmacMDBSanf 0.0000000
## EmmacMaclGeor 0.0000000 0.5212860

gl.dist.pop(gg, method="manhattan")

## Starting gl.dist.pop
## Processing a SNP dataset
## Calculating distances: manhattan
## Refer to dist {stats} documentation for algorithm

## Standard boxplot, no adjustment for skewness

35
 SNP data (DArTSeq)
Population manhattan Distances

0.0 0.1 0.2 0.3 0.4 0.5 0.6

15
10
5
0

0.0 0.1 0.2 0.3 0.4 0.5 0.6

##
## Reporting inter-population distances
## Distance measure: manhattan
## No. of populations = 7
## Average no. of individuals per population = 1
## No. of loci = 100
## Miniumum Distance: 0
## Maximum Distance: 0.58
## Average Distance: 0.135
##
## Completed: gl.dist.pop

## EmmacBurdMist EmmacBurnBara EmmacCoopEulb EmmacMDBForb

## EmmacBurnBara 0.0000000
## EmmacCoopEulb 0.0000000 0.0000000
## EmmacMDBForb 0.0000000 0.0000000 0.0000000
## EmmacMDBMaci 0.0000000 0.0000000 0.0000000 0.0000000
## EmmacMDBSanf 0.0000000 0.0000000 0.0000000 0.0000000
## EmmacMaclGeor 0.5681818 0.5747126 0.5681818 0.5813953
## EmmacMDBMaci EmmacMDBSanf
## EmmacBurnBara
## EmmacCoopEulb
## EmmacMDBForb
## EmmacMDBMaci
## EmmacMDBSanf 0.0000000
## EmmacMaclGeor 0.0000000 0.5434783

36
Distances available for computation include:
method = “manhattan”, “euclidean”, “canberra”, “bray”, “kulczynski”, “jaccard”, “gower”, “altGower”,
“morisita”, “horn”, “mountford”, “raup” , “binomial”, “chao”, “cao”, “mahalanobis”, “maximum”, “binary”
or “minkowski”.
Refer to the documentation for functions ?dist and ?vegdist for computational formulae.
YOu have been wondering why there are only 3 pairs of distances calculated. This is because gl.dist.pop
calculates distances using the population information in the genlight object. If someone wants to calculate
distances between individuals we need to redefine the population information in the genlight object. A
convinient way is to use the individuals identifier which is stored in the indNames slot.
Here is an example how to calculate individuals based distances for the first 7 individuals based on all loci:

glind7 <- gl[1:7,] #copy and store the original dataset in glind
pop(glind7) <- indNames(glind7)# redefine the population information

gl.dist.pop(glind7[1:7,], method="euclidean")

## Starting gl.dist.pop
## Processing a SNP dataset
## Calculating distances: euclidean
## Refer to dist {stats} documentation for algorithm

## Standard boxplot, no adjustment for skewness

SNP data (DArTSeq)

Population euclidean Distances

0.5 1.0 1.5 2.0 2.5

2.0
1.0
0.0

0.5 1.0 1.5 2.0 2.5

37
##
## Reporting inter-population distances
## Distance measure: euclidean
## No. of populations = 7
## Average no. of individuals per population = 1
## No. of loci = 255
## Miniumum Distance: 0.56
## Maximum Distance: 2.59
## Average Distance: 1.639
##
## Completed: gl.dist.pop

## AA010915 AA011723 AA012411 AA013214 AA019237 AA032760

## AA011723 1.8317377
## AA012411 2.5937207 2.0255179
## AA013214 1.3338177 2.0187477 2.4640269
## AA019237 1.7634501 0.9498014 2.0009056 1.6223072
## AA032760 1.0140147 1.5325417 2.4335429 0.5631733 1.4226066
## UC_00126 1.4577380 1.5473740 2.0814717 1.6758192 0.9475692 1.1464872

data.frame(ind=1:7, indNames=indNames(gl)[1:7], pop=pop(gl)[1:7])

## ind indNames pop

## 1 1 AA010915 EmmacMDBForb
## 2 2 UC_00126 EmmacMaclGeor
## 3 3 AA032760 EmmacMDBMaci
## 4 4 AA013214 EmmacMDBSanf
## 5 5 AA011723 EmmacBurnBara
## 6 6 AA012411 EmmacCoopEulb
## 7 7 AA019237 EmmacBurdMist

In the section below you will see that individuals from the area around the Cooper (individual 6, AA012411,
EmmacCoopEulb) are genetically quite different from the other populations. This is also reflected in the
largest euclidean pairwise distance for this individuals with every other individual (>2).

8.1 F Statistics

Several packages have already implemented the calculations of F statistics. Therefore we just provide different
ways how to convert our data sets and use already existing packages:
For Fst and Neis Gst we recommend the use of functions of the StAMPP package as they allow for the
use of parallel computing, which is much faster. It also works on genlight objects directly and therefore no
conversion is necessary. For a more detailed explanation on options to achieve also confidence intervals via
bootstrap refer to the StAMPP help pages) stamppFst calculates pairwise Fst values between populations.
(due to performance reasons we use only the first 30 individuals, which result in 8 populations :

library(StAMPP) #you may need to install the package

pwfst <-stamppFst(gl[1:20,], nboots=1, percent=95, nclusters=1)
round(pwfst,3)

For Neis Gst use

38
pwGst <-stamppNeisD(gl[1:20,]) #no parallel version :-()
round(pwGst,3)

For Jost’s D and G’st we suggest to use functions from the mmod package. The disadvantage here is that
it is much slower as we need to convert our data set from a genlight to a genind object and then use a
none-parallel function from mmod (depending on your computer the example will take some time).

library(mmod) #you may need to install the package first

#for performance reason use only a subset (and recode the populations)
recpops<- factor(rep(LETTERS[1:5],50))
glsub <- testset.gl
pop(glsub)<-recpops

gi <- gl2gi(glsub, v=0) #v=0 suppresses output

round(pairwise_D(gi),4)

## A B C D
## B -1e-04
## C 2e-04 -4e-04
## D 5e-04 -1e-04 2e-04
## E 0e+00 -4e-04 -1e-04 0e+00

round(pairwise_Gst_Hedrick(gi),4)

## A B C D
## B -0.0037
## C 0.0064 -0.0113
## D 0.0133 -0.0015 0.0064
## E -0.0011 -0.0118 -0.0028 -0.0014

round(pairwise_Gst_Nei(gi),4)

## A B C D
## B -0.0018
## C 0.0031 -0.0054
## D 0.0065 -0.0007 0.0031
## E -0.0005 -0.0057 -0.0014 -0.0007

9 Visualisation
Genetic similarity of individuals and populations can be visualized by way of Principal Coordinates Analysis
(PCoA) ordination (Gower, 1966). Individuals (entities) are represented in a space defined by loci (attributes)
with the position along each locus axis determined by genotype (0 for homozygous reference SNP, 2 for
homozygous alternate SNP, and 1 for the heterozygous state). Alternatively, populations can be regarded as
the entities to be plotted in a space defined by the loci, with the position along each locus axis determined
by the relative frequency of the alternate allele.
Orthogonal linear combinations of the original axes are calculated and ordinated such that the first PCoA
axis explains the most variation, PCoA-2 is orthogonal to PCoA-1 and explains the most residual variation,
and so on. A scree plot of eigenvalues provides an indication of the number of informative axes to examine,
viewed in the context of the average percentage variation explained by the original variables. The data are
typically presented in two or three dimensions in which emergent structure in the data is evident.

39
9.1 PCoA in dartR

The script gl.pcoa() is essentially a wrapper for glPca() of package adegenet with default settings apart
from setting parallel=FALSE, converting the eigenvalues to percentages and some additional diagnostics.

pc <- gl.pcoa(gl, nfactors=5)

## Starting gl.pcoa
## Processing a SNP dataset
## Performing a PCA, individuals as entities, loci as attributes, SNP genotype as state
Percentage Contribution

PCA on SNP Genotypes

Scree Plot (informative axes only)
20
5

2 4 6 8 10 12 14

PCA Axis

Noise Axes −− all eigenvalues positive

Frequency

15
0

0.000 0.005 0.010 0.015 0.020

Eigenvalue

## Completed: gl.pcoa

Please note, in case you are using a non-windows system you can use the argument ’‘’parallel=TRUE”’,
which speeds up the calculation. The resultant object pc contains the eigenvalues, factor scores and factor
loadings that can be accessed for subsequent analyses.

names(pc)

## [1] "scores" "eig" "loadings" "call"

The eigenvalues give the scaling factor for the eigenvectors (PCoA axis 1 - n), the scores give the coordinates
of the points (the entities, be they individuals or populations) in the new ordinated space, and the loadings

40
give the correlations of the original variables (the loci) against the new axes. Loci that load high on axis 1
are influential in discrimination among the entities in the direction of axis 1.
For example the percentage of variation the is represented by the axes can be calculated and visualised via:

barplot(pc$eig/sum(pc$eig)*100, )
20
15
10
5
0

9.2 Plotting the results of PCoA

The results of the PCoA can be plotted using gl.pcoa.plot() with a limited range of options. The script is
essentially a wrapper for plot {ggplot2} with the added functionality of {directlabels} and {plotly}.
The plotting script is not intended to produce publication quality plots, but should form a basis for importing
the plots to illustrator for subsequent amendment. The command

gl.pcoa.plot(pc, gl, labels="pop", xaxis=1, yaxis=2)

## Starting gl.pcoa.plot
## Plotting populations
## Preparing plot .... please wait

41
 EmvicVictJasp
EmmacJohnWari EmmacMDBForb EmmacRoss
EmmacClarYate
EmsubRopeMata EmmacMDBBowm
PCA Axis 2 ( 19.5 %)

0 EmmacMaclGeor EmmacMDBGwyd
EmmacNormSalt EmmacBurnBara EmmacMDBMaci EmmacMDBMurrM

EmmacMaryBoru EmmacMDBSanf EmmacMDBCond

EmmacNormJack EmmacTweeUki EmmacMDBCudg

−1
EmmacNormLeic EmmacBrisWive EmmacFitzAllig

EmmacBurdMist
EmmacCoopEulb
EmmacClarJack
−2 EmmacRichCasi
EmmacMaryPetr EmmacCoopAvin
−4 −3 −2 −1 0
PCA Axis 1 ( 23.3 %)
## While waiting, returning ggplot compliant object
## Completed: gl.pcoa.plot

42
 EmvicVictJasp
EmmacJohnWari EmmacMDBForb EmmacRoss
EmmacClarYate
EmsubRopeMata EmmacMDBBowm
PCA Axis 2 ( 19.5 %)

0 EmmacMaclGeor EmmacMDBGwyd
EmmacNormSalt EmmacBurnBara EmmacMDBMaci EmmacMDBMurrM

EmmacMaryBoru EmmacMDBSanf EmmacMDBCond

EmmacNormJack EmmacTweeUki EmmacMDBCudg

−1
EmmacNormLeic EmmacBrisWive EmmacFitzAllig

EmmacBurdMist
EmmacCoopEulb
EmmacClarJack
−2 EmmacRichCasi
EmmacMaryPetr EmmacCoopAvin
−4 −3 −2 −1 0
PCA Axis 1 ( 23.3 %)
You can see that this plot is very busy, and that the many labels are displaced quite some distance from their
associated points. This is because there is a tradeoff between avoiding overlap of the labels and proximity
of the labels - you can use colour to identify which labels go with which points. More sensibly, recoding
populations would be in order. We could use

glnew <- gl.edit.recode.pop(gl)

or using R

glnew <- testset.gl

levels(pop(glnew)) <- c(rep("Coast",5),rep("Cooper",3),rep("Coast",5),
rep("MDB",8),rep("Coast",7),"Em.subglobosa","Em.victoriae")
gl.pcoa.plot(pc, glnew, labels="pop", xaxis=1, yaxis=2)

## Starting gl.pcoa.plot
## Plotting populations
## Preparing plot .... please wait

43
 MDB

Em.victoriae
PCA Axis 2 ( 19.5 %)

0 Em.subglobosa

Coast

−1

Cooper

−2

−4 −3 −2 −1 0
PCA Axis 1 ( 23.3 %)
## While waiting, returning ggplot compliant object
## Completed: gl.pcoa.plot

44
 MDB

Em.victoriae
PCA Axis 2 ( 19.5 %)

0 Em.subglobosa

Coast

−1

Cooper

−2

−4 −3 −2 −1 0
PCA Axis 1 ( 23.3 %)
Note that we did not need to re run the PCoA analysis, only to recode the pop labels in the genlight object
that we hand to the plotting routine. Much clearer plot now.
There are other options for gl.pcoa.plot() that allow the axes to be scaled on the basis of proportion of
variation explained, to select other combinations of axes to plot, and for adding confidence ellipses. Use the
R help facility to explore these additional options.
Note that there is one point that seems intermediate between Emydura macquarii from the coast, and
Emydura subglobosa (from northern Australia west of the Great Dividing Range). How do we find out what
individual that point represents? Replot the data using labels=“interactive” to prime the plot for analysis
using ggplotly {plotly}:
In case ggplotly is not installed, please type:

install.packages("devtools")
library(devtools)
install_github("hadley/ggplot2")
library(ggplot2)

The commands

gl.pcoa.plot(pc, glnew, labels="interactive", xaxis=1, yaxis=2)

ggplotly()

will plot the individuals in the top two dimensions of the ordinated space, colour the points in accordance
to the population to which they belong, and allow points to be identified interactively using the mouse.

45
Now moving the mouse over the point reveals its identity. The animal is AA19157, from the coastal popu-
lations, and further scrutiny reveals it is from the Barron River in northern Queensland. Seems there has
been some allelic exchange there.

9.3 The Scree Plot

The number of dimensions with substantive information content can be determined by examining a scree
plot (Cattell, 1966).

gl.pcoa.scree(pc)

## Note: Only eigenvalues for dimensions that explain more that the average of the original variables
## No. of axes each explaining 10% or more of total variation: 3

## Warning: Use of ‘df$eigenvalue‘ is discouraged. Use ‘eigenvalue‘ instead.

## Warning: Use of ‘df$percent‘ is discouraged. Use ‘percent‘ instead.

## Warning: Use of ‘df$eigenvalue‘ is discouraged. Use ‘eigenvalue‘ instead.

## Warning: Use of ‘df$percent‘ is discouraged. Use ‘percent‘ instead.

46
 Percentage Contribution Scree plot for PCoA

20

15

10

0
0 5 10
Eigenvalue
This plot, by default, will show the percentage variation in the data explained by each axis successively where
the amount of variation is substantive. By substantive, I mean explaining more than the original variables
did on average. As a rule of thumb, one should examine all dimensions that explain more than 10% of the
variation in the data.

9.4 3D Plot

Should you find that 2 dimensions are insufficient to capture all substantive variation, you can examine a
plot of PCoA axis 2 against axis 1 and axis 3 against axis 1 and so on, taking care to note the proportion
of variation explained by each axis. Alternatively, when the data cluster tightly, additional dimensions can
be examined by removing all individuals from the analysis except those belonging to a single cluster and
re-running the PCoA (Georges and Adams, 1992). If three dimensions are indicated by the scree plot, as in
our current case, an interactive 3D plot can be produced

gl.pcoa.plot.3d(pc, glnew)

47
Note that the plot appears in a new window, outside R Studio, and that it is interactive in the sense that
you can rotate the plot using the mouse to obtain the most discriminatory view.
This function is essentially a wrapper for the corresponding function in {pca3d}, adding percentage variation
explained to each axis and fixing some parameters.

9.5 Neighbour-joining trees

Another way of visualizing genetic similarity among entities is to construct a phenetic tree using a neighbor-
joining approach or UPGMA. This script is essentially a wrapper for nj {ape} applied to Euclidean distances
between OTUs.

48
gl.tree.nj(glnew, type="fan")

## Starting gl.tree.nj
## Processing a SNP dataset
## Converting to a matrix of frequencies, locus by populations
## Computing Euclidean distances

Em
.su
bg
lob

st
Coa
MDB

Coo

ri
per

victo
Em.

## Completed: gl.tree.nj

##
## Phylogenetic tree with 5 tips and 3 internal nodes.
##
## Tip labels:
## [1] "Coast " "Cooper " "MDB " "Em.subglob" "Em.victori"
##
## Unrooted; includes branch lengths.

10 Fixed Difference Analysis

The PCoA approach outlined above considers allele frequency differences between individuals or populations
as a basis for constructing a distance matrix which is then used by the PCoA algorithms to execute the
ordination.

49
There is some advantage in considering only fixed differences between populations to examine structure, that
is, to use allelic differences where the alleles have come to fixation to alternative states in populations taken
pairwise.
A fixed difference between two populations at a specific locus occurs when the population share no alleles
at that locus. Allele frequencies may ebb and wane, but once a locus becomes fixed for an allele or suite
of alleles, there is no returning save a convergent mutation (rare) or gene flow. The acquistion of a fixed
difference between two populations is a significant biological event. Because it takes only one individual to
disperse between two populations per generation to retard divergence, the accumulation of fixed differences
between two populations is considered a robust indication of lack of gene flow.
In a nutshell, fixed differences are summed over populations taken pairwise, and when two populations have
no fixed differences (or insubstantial fixed differences), the populations are amalgamated and the process
repeated until there is no further reduction (Georges and Adams, 1996). The final set of taxa are diagnosable
by the presence or absence of a set of alleles at multiple loci.
As samples taken from populations for sequencing are finite, there are sampling issued to consider. There
is an asymmetry in these considerations. Populations can be amalgamated on the basis of shared alleles
at all loci with confidence, because it only takes one individual from one population to share alleles at a
locus with the other population to conlude that there is no fixed difference at that locus. So the process of
amalgamating populations based on shared alleles at all loci is robust.
This is not the case when interpeting fixed differences between two populations as a basis for their diag-
nosability. Because of the finite sample sizes, an observed fixed difference may have arisen because the
populations from which the samples were drawn do not share alleles at a particular locus, or because there
is a non-trivial probability that the observed fixed difference arose through sampling error (a false positive).
Distinguishing between real fixed differences and false positives is difficult, because we do not have knowledge
of the true allele fequencies for each of the two populations being compared. Our approach to dealing with
this is to turn to simulation. Briefly, we take the observed allele frequencies for the two populations under
consideration, take random samples of the allele profiles for each of the observed n1 and n2 individuals, count
the number of fixed differences, and repeat this, say 1000 times. This provides an estimate of the expected
number of fixed differences for the two populations.
Because the two populations may have true fixed differences, this estimate of expected fixed differences
conflates true fixed differences with false positives. We accommodate this by asking you to make a judgement.
How extreme does the divergence between the two populations have to be (say 99:1%, 1:99%) for you to
regard a sample fixed difference (100:0, 0:100) as a positive rather than a false positive? A value of delta =
0.02 (98:2, 2:98) might be a reasonable compromise. Then, any fixed differences that arise in the samples
from true allele frequencies of less extreme than 98:2, 2:98 are considered false positives. This decision
accommodates the possiblity of true fixed differences in the populations from which we draw the simulated
samples.
The script gl.collapse.recursive() optionally calculates the expected number of false positives (and its standard
error), and compares it with the observed number of fixed differences to yield a p value for the statistical
significance of the the result.
The script gl.collapse.pval() takes the final set amalgamated populations arising from gl.collapse.recursive(),
and further amalgamates populations based on lack of statistical significance of the observed fixed differences.
The resulting set of aggregated populations can be regarded as diagnostic OTUs for which you can be
confident there is strong support for lack of recent or contemporary gene flow between them, such that
phylogenetic approaches to extracting a bifrucating pattern of ancestry and descent are appropriate.
The script

fd <- gl.collapse.recursive(gl, t=0)

50
will ultimately yield a grouping of aggregate populations that are diagnosable from each other by one or
more fixed allelic differences. Here is the output, abridged: Let’s refresh the data again, to bring back in the
outgroups, then run the analysis.
The output comprises a list of matricies.
fdgl−−inputgenlightobject; f dfd – raw fixed differences; fdpcf d−−percentf ixeddif f erencesf dnobs – mean
no. of individuals used in each comparison fdnloc − −totalnumberof lociusedineachcomparisonf dexpobs
– if test=TRUE, the expected count of false positives for each comparison [by simulation] fd$prob – if
test=TRUE, the significance of the count of fixed differences [by simulation]
A full series of distance matrices and associated recode tables have been filed to disk for later interrogation
and use
An option to statistically compare the number of observed fixed differences with those expected to arise
by chance given the finite sample sizes is available using test=TRUE. Populations can then be further
amalgamated based on lack of significance using the script gl.collapse.pval().
The output is a list of matricies as above, but for the aggregated populations.
The outcome is unequivocal. All of the coastal Emydura macquarii and those from the Murray-Darling basin
amalgamate into one OTU on the basis of no fixed differences between them (Group 1.1). The Cooper Creek
animals form a separate diagnosable unit (OTU) (Group 1.2). The populations that did not amalgamate
outgroup taxa - Emydura victoriae and Emydura subglobosa - remain well supported.
You can see the power of this approach for defining diagnosable taxa, taxa that are free of contemporary or
recent geneflow and so on independent evolutionary trajectories.
##Distance Phylogeny on resultant OTUs
The script for distance phylogeny is gl2phylip() which calculates Euclidean distances using dist {stats} then
outputs the data in a form suitable for input to the Phylip package written by Joseph Felsenstein (http:
//evolution.genetics.washington.edu/phylip.html) (Felsenstein, 1989). The input file can include replicated
distance matrices for the purpose of bootstrapping.

gl <- fd.sig$gl
phy <- gl2phylip(gl, outfile="turtle.phy", bstrap=1000)

11 Isolation by distance plot

This functions performs an isolation by distance analysis based on a mantel test and also produces an
isolation by distance plot. If a genlight object with coordinates (in the @other$latlong slot) is provided,
then a Euclidean and genetic distance matrix are calculated (currently only pairwise Fst between population
is implemented by default). Coordinates are expected as lat long and converted to Google Earth Mercator
projections to avoid distortions. If coordinates are already projected in a suitable coordinate system, you
can set projected=TRUE and no projection is applied. An isolation by distance analysis will be carried out
and a plot will be returned, based on log(Euclidean distance) against between population pairwise Fst/1-Fst
(see Rousseau’s distance measure. Genetics April 1, 1997 vol. 145 no. 4 1219-1228) You can also provide
your own genetic and Euclidean distance matrix, then simply the mantel and isolation by distance plot
will be returned. The function is based on code provided by the adegenet tutorial (http://adegenet.r-forge.
r-project.org/files/tutorial-basics.pdf), using the functions mantel (package vegan), stamppFst (package
StAMPP) and Mercator (package dismo).
An example run (using only a subset of the test data):

gl <- gl.ibd(gl=testset.gl[1:180,])

51
## Standard analysis performed on the genlight object. Mantel test and plot will be Fst/1-Fst versus log
## Coordinates transformed to Mercator (google) projection to calculate distances in meters.
##
## Mantel statistic based on Pearson’s product-moment correlation
##
## Call:
## mantel(xdis = Dgen, ydis = Dgeo, permutations = 999, na.rm = TRUE)
##
## Mantel statistic r: 0.3655
## Significance: 0.001
##
## Upper quantiles of permutations (null model):
## 90% 95% 97.5% 99%
## 0.145 0.175 0.197 0.221
## Permutation: free
## Number of permutations: 999

Isolation by distance
4
3
Dgen

2
1
0

11 12 13 14

Dgeo

12 Phylogenetic Analysis
The objective of phylogenetic analysis is to extract relationships between taxonomic entities, be they species,
evolutionarily significant units (ESUs) or other diagnosable units (Felsenstein, 2004; Swofford and Berlocher,
1987). The goal is thus to extract the pattern of ancestry and descent among such taxonomic entities (call
them operational taxonomic units or OTUs). The OTUs need to be diagnosable because one assumes that
differences among them reflect divergence through time, unobscured by contemporary or recent tokogenic
exchange. That is, they are considered to be on evolutionary trajectories that are independent by virtue of
reproductive or long-standing geographic isolation.

52
The true evolutionary history of the OTUs is in the form of a bifurcating tree, that is, the true divergences
among OTUs satisfy both the conditions of a metric and the four-point condition (Buneman, 1973). Metric
is used here in the sense that, given the positions of two OTUs to represent the distance between them, the
distances to a third OTU uniquely defines its position. The four- point condition is used here in the sense
that for any four OTUs, there exists a simple tree accurately depicting the distances between them (that is,
there exists a non-negative internal branch). A set of OTUs and pairwise distances among them that satisfy
the metric and four-point conditions will define a unique bifurcating tree.
In practice, homoplasy obscures the true phylogeny by distorting measures of genetic distance between
taxa so that they no longer meet the metric and four-point criteria. The challenge becomes to estimate
the most parsimonious (least steps), most likely (maximum likelihood) or the best-bet (Bayesian estimate)
tree in terms of consistency with available data. Because information contained in the genetic code can be
over-written by new mutations, the true evolutionary history may not be recoverable even with the most
comprehensive of contemporary data. Instead, the exercise is to generate the phylogeny most consistent with
available data, and use this solution as the best available hypotheses for future testing, in whole or in part,
as new data or new methods of analyzing those data come to hand.
There are a number of approaches and a large range of software packages for recovering phylogenies which
are not covered here (Felsenstein, 1989; Swofford, 2002). The objective of {dartR} is to generate files in a
format that can be read into the packages of choice. We export to fastA format.

12.1 Distance Methods

Under ideal conditions, the true phylogeny can be uniquely recovered from the true measures of divergence
between OTUs (that is, tree distances define a unique tree). So one approach to recovering phylogenies
is to extract the phylogeny that is most consistent with the estimated distances between OTUs. To avoid
introducing artificial departure of the distances from the underlying tree, a metric distance needs to be
chosen, and for SNP datasets we choose Euclidean Distance. This differs from Rogers D (Rogers, 1972) by
a constant multiplier, and so the two measures are essentially the same.
The script for distance phylogeny is gl2phylip() which calculates Euclidean distances using dist {stats} then
outputs the data in a form suitable for input to the Phylip package written by Joseph Felsenstein (http:
//evolution.genetics.washington.edu/phylip.html) (Felsenstein, 1989). The input file can include replicated
distance matrices for the purpose of bootstrapping.
Assuming the data have been appropriately filtered before saving, the commands to do this are

gl <- testset.gl
phy <- gl2phylip(gl, outfile="turtle.phy", bstrap=1000)

The output file, turtle.phy or whatever you decide to call it, is available for input to Phylip and the variety
of commands for handling distance matricies (e.g. fitch). Refer to the Phylip documentation.

12.2 Character-based Methods

As with allozyme data sets of the past, SNP datasets present particular challenges for phylogenetic analysis.
The challenges arise because of difficulty in handling heterozygotes. Individuals homozygous for the reference
state (0) or homozygous for the alternate state (2) are unambiguous in their character state, but heterozygotes
(1) present both character states.

12.3 Converting Diploid to Haploid

One way of overcoming this is to randomly allocate the SNP state for heterozygous loci to one or the other
homozygous states, that is, to add background noise to the data from which the phylogenetic signal can

53
still be extracted without introducing a systematic bias. We convert diplotypes to haplotypes amenable to
phylogenetic analysis with gl2fasta() using method 1 to 4 (see ?gl2fasta for an explanation or the various
methods availalbe)
The way to do this is to use the command:

gl2fasta(gl, method=2, outfile="nohets.fasta")

which concatenates the sequences of the DNA fragments, trimmed of adaptors, into a fastA file, having first
randomly allocated the heterozygotic states. This set of “composite haplotypes” is then suitable for analysis
using the diverse range of phylogenetic packages available.
The reason for including the full trimmed sequences, most of the bases of which are monomorphic across
individuals, is to allow for application of various mutational models in likelihood analysis. These require
estimates of base frequencies and the frequency of transitions and transversions and the software to generate
these estimates and factor them in to the phylogenetic analysis e.g. ModelTest (Posada and Crandall 1998)
require the full sequence information.
The base frequencies and transition and transversion ratios can in any case be obtained using

gl.report.bases(testset.gl)

## Starting gl.report.bases
## Processing a SNP dataset
## Counting the bases
## Counting Transitions and Transversions
## Average trimmed sequence length: 60.7 ( 20 to 69 )
## Total number of trimmed sequences: 255
## Base frequencies (%)
## A: 25.01
## G: 26.67
## T: 24.12
## C: 24.2
##
## Transitions : 65.49
## Transversions: 34.51
## tv/ts ratio: 1.8977

54
 SNP: Base Frequencies

20
Frequency

10

0
A C G T
Bases

SNP: Ts/Tv Rates [ratio = 1.9 ]

60
Frequency

40

20

0
Ts Tv
Mutation Type

## Creating an output vector to return

## Completed: gl.report.bases

## $freq
## A G T C tv ts
## 25.01 26.67 24.12 24.20 34.51 65.49
##
## $plotbases

55
 SNP: Base Frequencies

20
Frequency

10

A C G T
Bases

##
## $plottstv

56
 SNP: Ts/Tv Rates [ratio = 1.9 ]

60

40
Frequency

20

Ts Tv
Mutation Type

Monomorphic sites are parsimony uninformative, so a more compact output fastA file, for parsimony analyses,
can be generated with

gl2fasta(gl, method=4, outfile="nohets.fasta")

12.4 Using Ambiguity Codes

Another common approach is to output the concatenated sequences using ambiguity codes for the heterozy-
gous SNP bases. This can be done with

gl2fasta(gl, method=1, outfile="ambcodes.fasta")

Maximum likelihood packages like RAxML Stamatakis (2014) cater for the ambiguities by adjusting tip
likelihoods as described by Felsenstein (2004:255). Again, it is possible to output only the variable sites to
the fastA file.

gl2fasta(gl, method=3, outfile="ambcodes.fasta")

These data can be used in some Maximum Likelihood software by providing base frequencies and transition
and transversion ratios separately. They can be calculated with

gl.report.bases(testset.gl)

57
## Starting gl.report.bases
## Processing a SNP dataset
## Counting the bases
## Counting Transitions and Transversions
## Average trimmed sequence length: 60.7 ( 20 to 69 )
## Total number of trimmed sequences: 255
## Base frequencies (%)
## A: 25.01
## G: 26.67
## T: 24.12
## C: 24.2
##
## Transitions : 65.49
## Transversions: 34.51
## tv/ts ratio: 1.8977

SNP: Base Frequencies

20
Frequency

10

0
A C G T
Bases

SNP: Ts/Tv Rates [ratio = 1.9 ]

60
Frequency

40

20

0
Ts Tv
Mutation Type

## Creating an output vector to return

## Completed: gl.report.bases

## $freq
## A G T C tv ts
## 25.01 26.67 24.12 24.20 34.51 65.49
##
## $plotbases

58
 SNP: Base Frequencies

20
Frequency

10

A C G T
Bases

##
## $plottstv

59
 SNP: Ts/Tv Rates [ratio = 1.9 ]

60

40
Frequency

20

Ts Tv
Mutation Type

##SVD Quartets
Concatenating the sequence tags or using data comprising of the SNP locus alone has been heavily criticised
in some quarters because the approach does not accommodate the possibility that the phylogeny for each
SNP may depart from the species tree. Gene trees need to be estimated separately and then combined to
estimate the species tree.
One such approach is to use SVD Quartets, that is, to estimate the relationships between taxa taken four
at a time, then combine these to estimate the species tree.
This is implemented in dartR using script gl2svdquartets(), which generates the nexus file for input to PAUP
for the analysis.
There are two approaches to preparing the data. The first enters the genotypes using two lines, one for
reference state, one for alternate state. The second approach enters one genotype per line, accommodating
the heterozygosity by using standard ambiguity codes. These are given as options in gl2svdquartets().
If using either of these approaches, it is important to filter out secondary SNPs from sequence tags using
gl.filter.secondaries(), prior to analysis.

13 Population Assignment
Assigning individuals of unknown provenance to populations of known provenance is a challenging exercise,
and several approaches have been suggested. Perhaps the simplest is to calculate the probabilities of yielding
the observed genotype of the unknown individual given the observed allele frequencies in each the target
population. Using this approach, the individual is assigned notionally to those populations for which this
probability is highest; populations for which the probability is lower than some level of significance are

60
eliminated from further consideration. This approach was first applied in a study of microsatellite markers
in bear populations (Paetkau et al. 2004) and subsequently applied using classical and Bayesian approaches
to estimating probabilities (Goetz and Thaller 1998; Blanchong et al. 2002).
Unfortunately, the sheer number of SNPs generated by next generation sequencing technologies, often in
the 10s or 100s of thousands, makes the assumption of independence of the loci untenable. Linkage is the
problem and for most organisms, there is insufficient genomic knowledge to overcome this directly. Similarly,
the sample sizes used in studies of population assignment are typically small, and inappropriate for tests of
sufficient power to identify loci that can be regarded as independent. Lack of independence is a problem
computationally because it is a prerequisite for combining the probabilities (or likelihoods) of the observed
genotype at each locus as a product (or by summation of log values) to yield an overall probability of
assignment for the unknown genotype. We have taken an alternate approach.
The approach taken here is to first eliminate from consideration those target populations where a SNP allele
is present in the unknown individual but not in the target. When the unknown individual possesses such a
private allele, the target population is unlikely to be the source population. This analysis can be done with
script gl.report.pa. The analysis is also undertaken by gl.assign.
In many cases, examining private alleles will narrow down the possible source populations considerably, and
depending on the spatial resolution required for the assignment (say, Australia or New Guinea), may provide
a satisfactory answer.
A second approach is to examine the position of the unknown individual relative to the target populations in
a reduced ordinated locus space using PCoA. This graphic representation is provided by gl.assign(). Addition
of confidence ellipses then allows a decision to eliminate some populations from consideration as the source
of the unknown individual.
The converse is not true. This approach does not allow assignment of the unknown to populations that
contain the unknown within their confidence ellipse. The overall confidence envelope is multidimensional,
and separation of the unknown from a target population may occur in deeper dimensions. Hence, as with
the private alleles approach, this graphical approach serves to narrow down the candidates for the source of
the unknown, and may in that sense, provide a satisfactory answer.
A third approach is to address the issue of non-independence (linkage) among the SNP loci by ordinating
the space defined by those loci. The resultant axes, linear combinations of the information contained in
each locus, are orthogonal and so can be regarded as independent. Subsequent standardization can achieve
independent and identically distributed variates, which simplifies analysis of probabilities and likelihoods.
The script gl.assign first eliminates populations on the basis of private alleles. It then ordinates the space
defined in locus space for the remaining populations.
A limited number of dimensions is retained in the final solution, the decision based on consideration of
(a) the number of substantive eigenvalues (greater in explanatory power than the original variables before
ordination), (b) the number of populations including the unknown, (c) an operational maximum number of
dimensions specified in the code (dim=7) or (d) a user specified value. The script selects the minimum of
these values to set the dimension of the reduced ordination space used subsequently.
A 95% confidence envelope (or some other level of confidence specified by the user) is defined in the reduced
ordinated space, and the likelihood of the unknown genotype occurring is estimated for each dimension
under Normal distribution assumptions. These likelihoods are logged for computational reasons, weighted
by the eigenvalue for their respective dimension, and summed to yield an Assignment Index for the unknown
against each population. Summing the weighted logged likelihoods is supported by the independence of each
of the ordinated axes, but the result should be nevertheless regarded as an assignment index rather than an
accurate likelihood.
An Assignment Index is calculated in the same way for a notional individual residing on the boundary of
the confidence envelope. Comparing the assignment index for each population with that of the notional
boundary individual provides a basis for a decision on assignment. If the Assignment Index for the unknown
is less than the critical value for the Assignment Index (that of the boundary individual), then the unknown

61
is assigned to that population. Where more than one population is selected, the population with the greatest
Assignment Index is the most likely. This will become evident in the examples that follow.

x <- gl.assign(testset.gl, unknown = "UC_00146", nmin=10, t=0)

## Starting gl.assign
## Processing a SNP dataset
## Starting utils.pa.ind
## Processing a SNP dataset
## Retaining 20 populations with sample size greater than or equal to 10 : EmmacBrisWive EmmacBurdMist
##
## Discarding 10 populations with sample size less than 10 : EmmacClarJack EmmacClarYate EmmacMDBGwyd
##
## Assigning 1 unknown individual(s) to 20 target populations
## Unknown individual: UC_00146
## Total number of SNP loci: 255
##
## Table showing number of loci with private alleles
## 0 EmmacMaclGeor
## 1 EmmacBurnBara EmmacRichCasi EmmacTweeUki
## 2 EmmacBrisWive EmmacMDBBowm EmmacMDBCudg EmmacMDBForb EmmacMDBMaci EmmacMDBMurrMung
## 3 EmmacFitzAllig EmmacMDBCond EmmacMDBSanf EmmacRussEube
## 4 EmmacBurdMist EmmacJohnWari EmmacRoss
## 6 EmmacCoopEulb
## 7 EmmacCoopCully
## 16 EmmacCoopAvin
##
## Data retained for the unknown individual and remaining candidate source populations (zero loci with
## EmmacMaclGeor unknown
##
## Completed: utils.pa.ind
## There are no further populations to compare for assignment. EmmacMaclGeor is the best assignment

In this example, the unknown individual, UC_00146, is assigned to only one population, EmmaclGeor, so
the assignment problem is resolved. This individual does indeed come from the Georges Creek, a tributary
of the Macleay River.
If we are worried that the 1 locus with a private allele out of 255 loci for populations EmmacBurnBara,
EmmacRichCasi, and EmmacTweeUki is less than substantial evidence, we can proceed to analyse the data
further using gl.assign. We would take this step also if there were more than one population scoring
0 private alleles. The code below yields the results of the private alleles analysis again, but in addition
computes assignment based on confidence envelopes:

x <- gl.assign(testset.gl, unknown ="UC_00146", nmin=10, alpha=0.95, t=1)

## Starting gl.assign
## Processing a SNP dataset
## Starting utils.pa.ind
## Processing a SNP dataset
## Retaining 20 populations with sample size greater than or equal to 10 : EmmacBrisWive EmmacBurdMist
##
## Discarding 10 populations with sample size less than 10 : EmmacClarJack EmmacClarYate EmmacMDBGwyd
##

62
## Assigning 1 unknown individual(s) to 20 target populations
## Unknown individual: UC_00146
## Total number of SNP loci: 255
##
## Table showing number of loci with private alleles
## 0 EmmacMaclGeor
## 1 EmmacBurnBara EmmacRichCasi EmmacTweeUki
## 2 EmmacBrisWive EmmacMDBBowm EmmacMDBCudg EmmacMDBForb EmmacMDBMaci EmmacMDBMurrMung
## 3 EmmacFitzAllig EmmacMDBCond EmmacMDBSanf EmmacRussEube
## 4 EmmacBurdMist EmmacJohnWari EmmacRoss
## 6 EmmacCoopEulb
## 7 EmmacCoopCully
## 16 EmmacCoopAvin
##
## Data retained for the unknown individual and remaining candidate source populations (zero loci with
## EmmacMaclGeor unknown
##
## Completed: utils.pa.ind
## There are no further populations to compare for assignment. EmmacMaclGeor is the best assignment

The plot allows us to eliminate populations EmmacRichCasi and EmmacBurnBara from consideration as
the unknown individual does not fall within or near their confidence ellipses. EmmacTweeUki and Emmac-
MaclGeor remain in contention.
Output from further analysis using the ordination and assignment index yields the following. Both Em-
macTweeUki and EmmacMaclGeor are potential sources for the unknown individual, but the most likely
source is EmmacMaclGeor, as supported by the private allele analysis.
A critical issue is whether the number of individuals in the target populations are sufficient to characterize
them in the critical considerations made here. Is the sample size sufficient to support the identification of
a private allele in the unknown? Is it sufficient to confidently construct confidence ellipses in the PCoA
plot? Is it sufficient to provide a robust estimate of the distribution of individuals along each axis of the
ordination in order to adequately estimate the likelihood of the unknown on that axis? We have set a default
of nmin=10, but this is a matter of judgement that needs to be considered when planning a study.

14 Sex Linkage
Sex linked SNP markers can obtained from genotypic profiles for known males and known females. Be aware
that the filters used by DArT Pty Ltd to generate the most reliable set of markers can apply tests that
eliminate sex linked markers and other markers that show abberrant patterns of inheritance. It is necessary
to ask DArT to generate a dataset fit for purpose, where they relax some of their filters.
Putative sex linked markers in a SNP dataset are those that are universally homozygous in the homogametic
sex and universally heterozygous in the heterogametic sex. The most promising putative sex linked markers
will have an AvgCount for the reference or alternate SNP of 10 or more, approximately one half in the other
SNP state.
The function gl.sexlinkage(gl) will identify these markers and provide diagnostics in the form of AvgCount.
The putative markers can then be tested more thoroughly by examining where they physically occur on
the chromosomes and by developing appropriate PCR tests that can be applied to the panel of known sex
individuals.

63
15 Conversion to formats for other packages
The genlight objects for managing SNP data are a relatively recent development, and the analysis options
are limited. Conversion to {adegenet} genind object opens up a greater range of analysis options, and this
conversion can be achieved with

gl <- testset.gl
gi <- gl2gi(gl, v=0)

Conversion in the opposite direction can be achieved with

gl2 <- gi2gl(gi)

15.1 Interfaces to Other Software

Package dartR does not pretend to provide a comprehensive range of analyses, but rather to provide avenues
from SNP data stored as a genlight object to other available software packages. The following conversion
functions are available:

function explanation
gl2fasta() Outputs the concatenated trimmed sequences to fastA format after first
converting heterozygous SNPs to ambiguity codes or randomly assigning the
heterozygous state to one or the other homozygous states (diplotypes to
haplotypes).
gl2genind() Converts a genlight object to a genind object as defined by the {adegenet}
package.
genind2gl() Converts a genind object as defined by the {adegenet} package to a genlight
object.
gl2nhyb() Outputs 200 loci selected according to user specified criteria for input to the
package NewHybrids (Anderson and Thompson, 2002).

gl2faststructure() gl2 gdsfmt() | Outputs a gl object to a file in gds format that can subsequently be used
with {SNPRelate}.

15.2 NewHybrids

NewHybrids (Anderson and Thompson, 2002) employs a statistical method for identifying species hybrids
using data on multiple, unlinked markers which does not require that allele frequencies be known in the
parental species. The probability model used is one in which parentals and various classes of hybrids (F1s,
F2s, and various backcrosses) form a mixture from which the sample is drawn. Using the framework of
Bayesian model-based clustering, NewHybrids computes, using Markov chain Monte Carlo, the relative like-
lihood (posterior probability) that each individual belongs to each of the distinct hybrid classes. NewHybrids
is limited to ca 200 loci because of memory constraints, so the best 200 available loci were selected on the
basis of their being different and fixed in each of the two nominated parental populations, and then on
avgPIC.
To run a NewHybrids analysis you need to have installed NewHybrids (http://ib.berkeley.edu/labs/slatkin/
eriq/software/software.htm#NewHybs)[http://ib.berkeley.edu/labs/slatkin/eriq/software/software.htm#
NewHybs].

64
glnew <- gl.nhybrids(testset.gl)

Refer to help (?gl.nhybrids) for further information. This function compares two sets of parental populations
to identify loci that exhibit a fixed difference, returns an genlight object with the reduced data, and creates
an input file for the program NewHybrids using the top 200 loci. In the absence of two identified parental
populations, the script will select a random set 200 loci only (method=random) or the first 200 loci ranked
on information content (AvgPIC).
The resultant NewHybrids input file is then passed to the program NewHybrids outside the R environment.
Refer to the NewHybrids documentation to continue.
If you have New Hybrids installed on your machine, then providing the directory with the New Hybrids
executable (e.g. C:/workspace/R_analysis/NewHybsPC) will result in the analysis being executed. The
advantage of this is that the final output is a little more elaborative.

15.3 Phylip

PHYLIP (Felsenstein, 1989) is a package of programs for inferring phylogenies (evolutionary trees). Methods
that are available in the package include parsimony, distance matrix, and likelihood methods, including
bootstrapping and consensus trees. The {dartR} scripts produce Euclidean distance matrices in a form that
can be input to Phylip distance analyses (e.g. program Fitch).
To generate an input matrix for Phylip, use the function

glnew <- gl2phylip(outfile = "phyinput.txt")

To generate an input file containing resampled input matrices for the purposes of bootstrapping, use:

gl.new <- gl2phylip(outfile = "phyinput.txt", bstrap = 1000)

The resultant output file can be passed to Phylip programs for execution. Refer to the Phylip documentation.

15.4 SNPRelate

R package SNPRelate is available to undertake principal components analysis and relatedness analysis (Zheng
et al., 2012). {SNPRelate} expects the data to be in gds format. As with the genlight format of package
{adegenet}, the gds format supports efficient memory management and storage of SNP genotypes and
associated metadata. In this format each byte encodes up to four SNP genotypes thereby reducing file size
and access time. The GDS format supports data blocking so that only the subset of data that is being
processed needs to reside in memory. GDS formatted data is also designed for efficient random access to
large data sets.
Data conversion from a genlight object to gds format is via

gl2gds(gl, outfile="test.gds")

Once this file is created, you can open it for analysis with {SNPRelate} using

gds <- snpgdsOpen("gl2gds.gds")

and undertake the range of analyses available in {SNPRelate}, including

65
 option explanation
LD-pruning A pruned set of loci which are in approximate linkage equilibrium will avoid
the strong influence of locus clusters on relatedness analysis
PCoA with some nice diagnostic plots
Relatedness Identity-by-descent estimation can be done by either the method of
moments (MoM) or maximum likelihood estimation (MLE)
IBS Identity by State analysis and Multidimensional Scaling (MDS) for
displaying relationships

15.5 FastSTRUCTURE

STRUCTURE is one of the most widely used population analysis tools that allows researchers to assess
patterns of genetic structure in a set of samples (Porras-Hurtado et al., 2013). STRUCTURE is freely
available software for population analysis (Pritchard et al., 2000). STRUCTURE analyses differences in
the distribution of genetic variants amongst populations and places individuals into groups where they share
similar patterns of variation. STRUCTURE both identifies populations from the data and assigns individuals
to those populations. FastSTRUCTURE is an improved implementation to analyse large quantities of data
(Raj et al., 2014).
To generate an input file for fastSTRUCTURE, use the function (this format can also be used for input to
STRUCTURE, though the meta data options are not supported yet):

gl2faststructure(gl, outfile="myfile.fs", probar = FALSE)

## Starting gl2faststructure
## Processing a SNP dataset
## Saved faststructure file: C:\Users\s425824\AppData\Local\Temp\Rtmp61wFlf/myfile.fs
## Completed: gl2faststructure

## NULL

16 References
Anderson, E.C., Thompson, E.A., 2002. A Model-Based Method for Identifying Species Hybrids Using
Multilocus Genetic Data. Genetics 160, 1217-1229.
Blanchong, J. A., Scribner, K. T., and Winterstein, S. R. (2002). Assignment of Individuals to Populations:
Bayesian Methods and Multi-Locus Genotypes. The Journal of Wildlife Management 66, 321. doi:10.2307/
3803164
Buneman, P., 1973. A note on the metric properies of trees. J Combinarorial Theory 17B, 48-50.
Cattell, R.B., 1966. The scree test for the number of factors. Multivariate Behavioral Research 1, 245-276.
doi:10.1207/s15327906mbr0102_10
Felsenstein, J., 2004. Inferring Phylogenies. Sinauer Associates, Sunderdland, MA USA.
Felsenstein, J., 1989. PHYLIP - Phylogeny Inference Package (Version 3.2). Cladistics 5, 164-166.
Georges, A., Adams, M., 1996. Electrophoretic delineation of species boundaries within the short-necked
freshwater turtles of Australia (Testudines: Chelidae). Zoological Journal of the Linnean Society 118, 241-
260.

66
Georges, A., Adams, M., 1992. A phylogeny for Australian chelid turtles based on allozyme electrophoresis.
Australian Journal of Zoology 40, 453-476.
Götz, K.-U., and Thaller, G. (1998). Assignment of individuals to populations using microsatellites. Journal
of Animal Breeding and Genetics 115, 53–61.
Gower, J.C., 1966. Some distance properties of latent root and vector methods used in multivariate analysis.
Biometrika 53, 325–338. doi:10.1093/biomet/53.3-4.325
Jombart, T., 2008. adegenet: a R package for the multivariate analysis of genetic markers. Bioinformatics
24, 1403–1405. doi:10.1093/bioinformatics/btn129
Jombart, T., Ahmed, I., 2011. adegenet 1.3-1: new tools for the analysis of genome-wide SNP data. Bioin-
formatics 27, 3070–3071. doi:10.1093/bioinformatics/btr521
Paetkau, D., Slade, R., Burden, M., and Estoup, A. (2004). Genetic assignment methods for the direct,
real-time estimation of migration rate: a simulation-based exploration of accuracy and power. Molecular
Ecology 13, 55–65.
Porras-Hurtado, L., Ruiz, Y., Santos, C., Phillips, C., Carracedo, Á., Lareu, M.V., 2013. An overview of
STRUCTURE: applications, parameter settings, and supporting software. Frontiers in Genetics 4. doi:
10.3389/fgene.2013.00098
Pritchard, J.K., Stephens, M., Donnelly, P., 2000. Inference of population structure using multilocus geno-
type data. Genetics 155, 945–959.
Porras-Hurtado, L., Ruiz, Y., Santos, C., Phillips, C., Carracedo, L., Lareu, M.V., 2013. An overview of
STRUCTURE: applications, parameter settings, and supporting software. Frontiers in Genetics 4. doi:
10.3389/fgene.2013.00098
Raj, A., Stephens, M., Pritchard, J.K., 2014. fastSTRUCTURE: Variational Inference of Population Struc-
ture in Large SNP Data Sets. Genetics 197, 573–589. doi:10.1534/genetics.114.164350
Rogers, J.S., 1972. Measures of similarity and genetic distance. Studies in Genetics 7, 145–153.
Swofford, D.L., 2002. Phylogenetic Analysis Using Parsimony (*and Other Methods). Version 4. Sinauer
Associates, Sunderland, Massachusetts, USA.
Stamatakis, A., 2014. RAxML version 8: a tool for phylogenetic analysis and post-analysis of large phylo-
genies. Bioinformatics, 30(9), 1312-1313.
Swofford, D.L., Berlocher, S.H., 1987. Inferring evolutionary trees from gene frequency data under the
principle of maximum parsimony. Systematic Zoology 36, 293–325.
Zheng, X., Levine, D., Shen, J., Gogarten, S.M., Laurie, C., Weir, B.S., 2012. A high-performance computing
toolset for relatedness and principal component analysis of SNP data. Bioinformatics 28, 3326–3328.

67