dapc: Discriminant Analysis of Principal Components (DAPC)

Description

These functions implement the Discriminant Analysis of Principal Components (DAPC, Jombart et al. 2010). This method descibes the diversity between pre-defined groups. When groups are unknown, use find.clusters to infer genetic clusters. See 'details' section for a succint description of the method, and vignette("adegenet-dapc") for a tutorial. Graphical methods for DAPC are documented in scatter.dapc (see ?scatter.dapc).

dapc is a generic function performing the DAPC on the following types of objects: - data.frame (only numeric data) - matrix (only numeric data) - genind objects (genetic markers) - genlight objects (genome-wide SNPs) These methods all return an object with class dapc.

Functions that can be applied to these objects are (the ".dapc" can be ommitted):

- print.dapc: prints the content of a dapc object. - summary.dapc: extracts useful information from a dapc object. - predict.dapc: predicts group memberships based on DAPC results.

DAPC implementation calls upon dudi.pca from the ade4 package (except for genlight objects) and lda from the MASS package. The predict procedure uses predict.lda from the MASS package.

as.lda is a generic with a method for dapc object which converts these objects into outputs similar to that of lda.default.

Usage

## S3 method for class 'data.frame':
dapc(x, grp, n.pca=NULL, n.da=NULL, center=TRUE,
     scale=FALSE,var.contrib=TRUE, pca.info=TRUE, pca.select=c("nbEig","percVar"),
     perc.pca=NULL, ..., dudi=NULL)
## S3 method for class 'matrix':
dapc(x, \ldots)
## S3 method for class 'genind':
dapc(x, pop=NULL, n.pca=NULL, n.da=NULL, scale=FALSE,
     scale.method=c("sigma", "binom"), truenames=TRUE, var.contrib=TRUE,
     pca.info=TRUE, pca.select=c("nbEig","percVar"), perc.pca=NULL, ...)
## S3 method for class 'genlight':
dapc(x, pop = NULL, n.pca = NULL, n.da = NULL, scale
    = FALSE, var.contrib = TRUE, pca.info=TRUE, pca.select = c("nbEig", "percVar"),
    perc.pca = NULL, glPca = NULL, ...)
## S3 method for class 'dudi':
dapc(x, grp, \ldots)
## S3 method for class 'dapc':
print(x, \dots)
## S3 method for class 'dapc':
summary(object, \dots)
## S3 method for class 'dapc':
predict(object, newdata, prior = object$prior, dimen,
         method = c("plug-in", "predictive", "debiased"), ...)

Arguments

a data.frame, matrix, or genind object. For the data.frame and matrix arguments, only quantitative variables should be provided.

grp,pop

a factor indicating the group membership of individuals; for scatter, an optional grouping of individuals.

n.pca

an integer indicating the number of axes retained in the Principal Component Analysis (PCA) step. If NULL, interactive selection is triggered.

n.da

an integer indicating the number of axes retained in the Discriminant Analysis step. If NULL, interactive selection is triggered.

center

a logical indicating whether variables should be centred to mean 0 (TRUE, default) or not (FALSE). Always TRUE for genind objects.

scale

a logical indicating whether variables should be scaled (TRUE) or not (FALSE, default). Scaling consists in dividing variables by their (estimated) standard deviation to account for trivial differences in variances. Further scalin

var.contrib

a logical indicating whether the contribution of original variables (alleles, for genind objects) should be provided (TRUE, default) or not (FALSE). Such output can be useful, but can also create huge ma

pca.info

a logical indicating whether information about the prior PCA should be stored (TRUE, default) or not (FALSE). This information is required to predict group membership of new individuals using predict, but makes the ob

pca.select

a character indicating the mode of selection of PCA axes, matching either "nbEig" or "percVar". For "nbEig", the user has to specify the number of axes retained (interactively, or via n.pca). For "percVar", the user h

perc.pca

a numeric value between 0 and 100 indicating the minimal percentage of the total variance of the data to be expressed by the retained axes of PCA.

...

further arguments to be passed to other functions. For dapc.matrix, arguments are to match those of dapc.data.frame; for dapc.genlight, arguments passed to glPca

glPca

an optional glPca object; if provided, dimension reduction is not performed (saving computational time) but taken directly from this object.

object

a dapc object.

scale.method

a character specifying the scaling method to be used for allele frequencies, which must match "sigma" (usual estimate of standard deviation) or "binom" (based on binomial distribution). See scale

truenames

a logical indicating whether true (i.e., user-specified) labels should be used in object outputs (TRUE, default) or not (FALSE).

dudi

optionally, a multivariate analysis with the class dudi (from the ade4 package). If provided, prior PCA will be ignored, and this object will be used as a prior step for variable orthogonalisation.

newdata

an optional dataset of individuals whose membership is seeked; can be a data.frame, a matrix, a genind or a genlight object, but object class must match the original ('training') data. In p

prior,dimen,method

see ?predict.lda.

Value

=== dapc objects === The class dapc is a list with the following components:
callthe matched call.
n.pcanumber of PCA axes retained
n.danumber of DA axes retained
varproportion of variance conserved by PCA principal components
eiga numeric vector of eigenvalues.
grpa factor giving prior group assignment
priora numeric vector giving prior group probabilities
assigna factor giving posterior group assignment
tabmatrix of retained principal components of PCA
loadingsprincipal axes of DAPC, giving coefficients of the linear combination of retained PCA axes.
ind.coordprincipal components of DAPC, giving the coordinates of individuals onto principal axes of DAPC; also called the discriminant functions.
grp.coordcoordinates of the groups onto the principal axes of DAPC.
posteriora data.frame giving posterior membership probabilities for all individuals and all clusters.
var.contr(optional) a data.frame giving the contributions of original variables (alleles in the case of genetic data) to the principal components of DAPC.
=== other outputs === Other functions have different outputs: - summary.dapc returns a list with 6 components: n.dim (number of retained DAPC axes), n.pop (number of groups/populations), assign.prop (proportion of overall correct assignment), assign.per.pop (proportion of correct assignment per group), prior.grp.size (prior group sizes), and post.grp.size (posterior group sizes).

encoding

UTF-8

Details

The Discriminant Analysis of Principal Components (DAPC) is designed to investigate the genetic structure of biological populations. This multivariate method consists in a two-steps procedure. First, genetic data are transformed (centred, possibly scaled) and submitted to a Principal Component Analysis (PCA). Second, principal components of PCA are submitted to a Linear Discriminant Analysis (LDA). A trivial matrix operation allows to express discriminant functions as linear combination of alleles, therefore allowing one to compute allele contributions. More details about the computation of DAPC are to be found in the indicated reference.

DAPC does not infer genetic clusters ex nihilo; for this, see the find.clusters function.

References

Jombart T, Devillard S and Balloux F (2010) Discriminant analysis of principal components: a new method for the analysis of genetically structured populations. BMC Genetics11:94. doi:10.1186/1471-2156-11-94

Examples

Run this code

## data(dapcIllus), data(eHGDP), and data(H3N2) illustrate the dapc
## see ?dapcIllus, ?eHGDP, ?H3N2
##
example(dapcIllus)
example(eHGDP)
example(H3N2)

## H3N2 EXAMPLE ##
data(H3N2)
pop(H3N2) <- factor(H3N2$other$epid)
dapc1 <- dapc(H3N2, var.contrib=FALSE, scale=FALSE, n.pca=150, n.da=5)

## remove internal segments and ellipses, different pch, add MStree
scatter(dapc1, cell=0, pch=18:23, cstar=0, mstree=TRUE, lwd=2, lty=2)

## only ellipse, custom labels
scatter(dapc1, cell=2, pch="", cstar=0, posi.da="top",
lab=paste("year
",2001:2006), axesel=FALSE, col=terrain.colors(10))


## SHOW COMPOPLOT ON MICROBOV DATA ##
data(microbov)
dapc1 <- dapc(microbov, n.pca=20, n.da=15)
compoplot(dapc1, lab="")




## EXAMPLE USING GENLIGHT OBJECTS ##
## simulate data
x <- glSim(50,4e3-50, 50, ploidy=2)
x
plot(x)

## perform DAPC
dapc1 <- dapc(x, n.pca=10, n.da=1)
dapc1

## plot results
scatter(dapc1, scree.da=FALSE)

## SNP contributions
loadingplot(dapc1$var.contr)
loadingplot(tail(dapc1$var.contr, 100), main="Loading plot - last 100 SNPs")



## USE "PREDICT" TO PREDICT GROUPS OF NEW INDIVIDUALS ##
## load data
data(sim2pop)

## we make a dataset of:
## 30 individuals from pop A
## 30 individuals from pop B
## 30 hybrids

## separate populations and make F1
temp <- seppop(sim2pop)
temp <- lapply(temp, function(e) hybridize(e,e,n=30)) # force equal popsizes

## make hybrids
hyb <- hybridize(temp[[1]], temp[[2]], n=30)

## repool data - needed to ensure allele matching
newdat <- repool(temp[[1]], temp[[2]], hyb)
pop(newdat) <- rep(c("pop A", "popB", "hyb AB"), c(30,30,30))

## perform the DAPC on the first 2 pop (60 first indiv)
dapc1 <- dapc(newdat[1:60],n.pca=5,n.da=1)

## plot results
scatter(dapc1, scree.da=FALSE)

## make prediction for the 30 hybrids
hyb.pred <- predict(dapc1, newdat[61:90])
hyb.pred

## plot the inferred coordinates (circles are hybrids)
points(hyb.pred$ind.scores, rep(.1, 30))

## look at assignment using assignplot
assignplot(dapc1, new.pred=hyb.pred)
title("30 indiv popA, 30 indiv pop B, 30 hybrids")

## image using compoplot
compoplot(dapc1, new.pred=hyb.pred, ncol=2)
title("30 indiv popA, 30 indiv pop B, 30 hybrids")

Run the code above in your browser using DataLab

Data Engineering and BI courses are free this week!