scaleGen-methods {adegenet} | R Documentation |
The generic function scaleGen
is an analogue to the
scale
function, but is designed with further arguments giving
scaling options.
Methods are defined for genind and genpop objects. Both return data.frames of scaled allele frequencies.
## S4 method for signature 'genind': scaleGen(x, center=TRUE, scale=TRUE, method=c("sigma", "binom"), missing=c("NA","0","mean"),truenames=TRUE) ## S4 method for signature 'genpop': scaleGen(x, center=TRUE, scale=TRUE, method=c("sigma", "binom"), missing=c("NA","0","mean"),truenames=TRUE)
x |
a genind and genpop object |
center |
a logical stating whether alleles frequencies should be centred to mean zero (default to TRUE). Alternatively, a vector of numeric values, one per allele, can be supplied: these values will be substracted from the allele frequencies. |
scale |
a logical stating whether alleles frequencies should be scaled (default to TRUE). Alternatively, a vector of numeric values, one per allele, can be supplied: these values will be substracted from the allele frequencies. |
method |
a character indicating the method to be used. See details. |
truenames |
a logical indicating whether true labels (as opposed to generic labels) should be used to name the output. |
missing |
a character giving the treatment for missing values. Can be "NA", "0" or "mean" |
The argument method
is used as follows:
- sigma
: scaling is made using the usual standard deviation
- binom
: scaling is made using the theoretical variance of the
allele frequency. This can be used to avoid that frequencies close to
0.5 have a stronger variance that those close to 0 or 1.
A matrix of scaled allele frequencies with genotypes (genind) or populations in (genpop) in rows and alleles in columns.
Thibaut Jombart jombart@biomserv.univ-lyon1.fr
## load data data(microbov) obj <- genind2genpop(microbov) ## compare different scaling X1 <- scaleGen(obj) X2 <- scaleGen(obj,met="bin") if(require(ade4)){ ## compute PCAs pcaObj <- dudi.pca(obj,scale=FALSE,scannf=FALSE) # pca with no scaling pcaX1 <- dudi.pca(X1,scale=FALSE,scannf=FALSE,nf=100) # pca with usual scaling pcaX2 <- dudi.pca(X2,scale=FALSE,scannf=FALSE,nf=100) # pca with scaling for binomial variance ## get the loadings of alleles for the two scalings U1 <- pcaX1$c1 U2 <- pcaX2$c1 ## find an optimal plane to compare loadings ## use a procustean rotation of loadings tables pro1 <- procuste(U1,U2,nf=2) ## graphics par(mfrow=c(2,2)) # eigenvalues barplot(pcaObj$eig,main="Eigenvalues\n no scaling") barplot(pcaX1$eig,main="Eigenvalues\n usual scaling") barplot(pcaX2$eig,main="Eigenvalues\n 'binomial' scaling") # differences between loadings of alleles s.match(pro1$scor1,pro1$scor2,clab=0,sub="usual -> binom (procustean rotation)") }