processDatasetAllo runs alleleCorrelations on every locus in
a "genambig" object, then runs testAlGroups on every locus
using several user-specified parameter sets. It chooses a single best set of allele assignments
for each locus, and produces plots to help the user evaluate assignment quality.
plotSSAllo assists the user in evaluating the quality of allele assignments by plotting
the results of K-means clustering. plotParamHeatmap assists the user in choosing the
best parameter set for testAlGroups for each locus.
plotSSAllo(AlCorrArray)
plotParamHeatmap(propMat, popname = "AllInd", col = grey.colors(12)[12:1], main = "")
processDatasetAllo(object, samples = Samples(object), loci = Loci(object),
n.subgen = 2, SGploidy = 2, n.start = 50, alpha = 0.05,
parameters = data.frame(tolerance = c(0.05, 0.05, 0.05, 0.05),
swap = c(TRUE, FALSE, TRUE, FALSE),
null.weight = c(0.5, 0.5, 0, 0)),
plotsfile = "alleleAssignmentPlots.pdf", usePops = FALSE, ...)plotSSAllo draws a plot and invisibly returns a list:
A two-dimensional array with loci in the first dimension and populations in the second dimension. Each value is the sums-of-squares between isoloci divided by the total sums-of-squares, as output by K-means clustering. If K-means clustering was not performed, the value is zero.
An array of the same dimensions as $ssratio, containing values to indicate how evenly
alleles are distributed among isoloci as determined by K-means clustering. This is:
$$1 - \sum_{1}^{i}{(\frac{a_{i}}{A})^2}$$
where \(i\) is the number of isoloci, \(a_i\) is the number of alleles for a given isolocus, and \(A\)
is the total number of alleles for the locus.
The maximum possible value for $evenness, given the number of isoloci.
The minimum possible value for $evenness, given the number of isoloci and alleles.
An array of the same dimensions as $ssratio, containing TRUE if there were any
positive correlations between alleles, and FALSE if not.
processDatasetAllo returns a list:
A two-dimensional list with loci in the first dimension and populations in the second
dimension, giving the results of alleleCorrelations.
A three-dimensional list with loci in the first dimension, populations in the second dimension,
and parameter sets in the third dimension, giving the results of testAlGroups.
The output of plotSSAllo.
A three-dimensional array, with the same dimensions as $TAGarray, indicating
the proportion of alleles that were found to be homoplasious for each locus, population, and parameter set.
A two-dimensional list, with loci in the first dimension and parameter sets in the
second dimension, containing allele assignments merged across populations. This is the output of
mergeAlleleAssignments.
A two-dimensional array, of the same dimensions as $mergedAssignments, indicating
the proportion of alleles that were homoplasious, for each locus and parameter set, for allele assignments that
were merged across populations.
A matrix with the same dimensions as $mergedAssignments indicating the proportion of
non-missing genotypes from the original dataset that cannot be unambiguously recoded, without invoking
aneuploidy, using the merged allele assignments from each parameter set for each locus.
A one-dimensional list with a single best set of allele assignments, from $mergedAssignments,
for each locus. The best set of assignments is chosen using $missRate, then in the case of a tie using
$propHomoplMerged, then in the case of a tie using the parameter set that was listed first.
plotParamHeatmap draws a plot and does not return anything.
A two-dimensional list, where each item in the list is the output of alleleCorrelations.
The first dimension represents loci, and the second dimension represents populations. Both
dimensions are named. This is the $AlCorrArray output of processDatasetAllo.
A two-dimensional array, with loci in the first dimension and parameter sets in the
second dimension, indicating the proportion of alleles that were found to be homoplasious by testAlGroups
or the proportion of genotypes that could not be recoded using a given set of allele assignments.
This can be the $propHomoplasious output of processDatasetAllo, indexed by a single population. If a
three-dimensional array is provided, it will be indexed in the second dimension by popname. The
$propHomoplMerged or $missRate output of processDatasetAllo may also be passed to this argument.
The name of the population corresponding to the data in propMat.
The color scale for representing the proportion of loci that are homoplasious or the proportion of genotypes that are missing.
A title for the plot.
A "genambig" object.
An optional character vector indicating which samples to include in analysis.
An optional character vector indicating which loci to include in analysis.
The number of isoloci into which each locus should be split. Passed directly to
alleleCorrelations.
The ploidy of each isolocus. Passed directly to testAlGroups.
Passed directly to the nstart argument of kmeans. See alleleCorrelations.
The significance threshold for determining whether two alleles are significantly correlated. Used
primarily for identifying potentially problematic positive correlations. Passed directly to alleleCorrelations.
Data frame indicating parameter sets to pass to testAlGroups. Each row is one set of parameters.
A PDF output file name for drawing plots to help assess assignment quality. Can be NULL if no
plots are desired.
If TRUE, population assignments are taken from the PopInfo slot of object, and
populations are analyzed separately with alleleCorrelations and testAlGroups, before merging the
results with mergeAlleleAssignments.
Additional parameters to pass to testAlGroups for adjusting the simulated annealing algorithm.
Lindsay V. Clark
plotSSAllo produces a plot of loci by population, with the sums-of-squares ratio on the x-axis and the evenness of allele distribution
on the y-axis (see Value). Locus names are written directly on the plot. If there are multiple population names, locus names are colored
by population, and a legend is provided for colors. Loci with high-quality allele clustering are expected to be in the upper-right
quadrant of the plot. If locus names are in italics, it indicates that positive correlations were found between some alleles, indicating
population structure or scoring error that could interfere with assignment quality.
plotParamHeatmap produces an image to indicate the proportion of alleles found to be homoplasious, or the proportion of genotypes that
could not be unambiguously recoded using allele assignments, for each locus and
parameter set for a given population (when looking at homoplasy) or merged across populations (for homoplasy or the proportion of non-recodeable
genotypes). Darker colors indicate more homoplasy or more genotypes that could not be recoded.
The word “best” indicates, for each
locus, the parameter set that found the least homoplasy or smallest number of non-recodeable genotypes.
By default, processDatasetAllo generates a PDF file containing output from plotSSAllo and plotParamHeatmap,
as well as heatmaps of the $heatmap.dist output of alleleCorrelations for each locus and population.
Heatmaps are not plotted for loci where an allele is present in all individuals. processDatasetAllo also
generates a list of R objects containing allele assignments under different parameters, as well as statistics for evaluating
clustering quality and choosing the optimal parameter sets, as described below.
Clark, L. V. and Drauch Schreier, A. (2017) Resolving microsatellite genotype ambiguity in populations of allopolyploid and diploidized autopolyploid organisms using negative correlations between allelic variables. Molecular Ecology Resources, 17, 1090--1103. DOI: 10.1111/1755-0998.12639.
alleleCorrelations, recodeAllopoly
# get example dataset
data(AllopolyTutorialData)
# data cleanup
mydata <- deleteSamples(AllopolyTutorialData, c("301", "302", "303"))
PopInfo(mydata) <- rep(1:2, each = 150)
Genotype(mydata, 43, 2) <- Missing(mydata)
# allele assignments
# R is set to 10 here to speed processing for example. It should typically be left at the default.
myassign <- processDatasetAllo(mydata, loci = c("Loc3", "Loc6"),
plotsfile = NULL, usePops = TRUE, R = 10,
parameters = data.frame(tolerance = c(0.5, 0.5),
swap = c(TRUE, FALSE),
null.weight = c(0.5, 0.5)))
# view best assignments for each locus
myassign$bestAssign
# plot K-means results
plotSSAllo(myassign$AlCorrArray)
# plot proportion of homoplasious alleles
plotParamHeatmap(myassign$propHomoplasious, "Pop1")
plotParamHeatmap(myassign$propHomoplasious, "Pop2")
plotParamHeatmap(myassign$propHomoplMerged, "Merged across populations")
# plot proportion of missing data, after recoding, for each locus and parameter set
plotParamHeatmap(myassign$missRate, main = "Missing data:")
Run the code above in your browser using DataLab