Combine several dissimilarity distance matrices

Source: R/PRE_FATE.speciesDistanceCombine.R

This script is designed to create a distance matrix between species, combining several dissimilarity distance matrices.

PRE_FATE.speciesDistanceCombine(
  list.mat.dist,
  opt.min.noMat = length(list.mat.dist),
  opt.normal = TRUE,
  opt.weights = NULL
)

Arguments

list.mat.dist

a list of matrices containing dissimilarity distance values between each pair of species.

opt.min.noMat

(optional) default length(list.mat.dist).
An integer corresponding to the minimal number of distance matrices for which each species should have values

opt.normal

(optional) default TRUE.
If TRUE, all given distance matrices will be normalized (see Details)

opt.weights

(optional) default NULL.
A vector of double (between 0 and 1) corresponding to the weights for each distance matrix provided in list.mat.dist. They must sum up to 1.

Value

A matrix containing the weighted (or not) combination of the different transformed (or not) distance matrices given.


The information for the combination of all distances is written in PRE_FATE_DOMINANT_speciesDistance.csv file.

Details

This function allows to obtain a distance matrix between species, based on several dissimilarity distance matrices combined according to the following formula :

$$\text{mat.DIST} = \Sigma (\text{wei.i} * \text{mat.DIST}_{i})$$

If opt.normal = TRUE, two normalization steps are applied to each distance matrix before combining them :

  1. a non-paranormal (npn) transformation (huge.npn function) to obtain Gaussian distributions for all dissimilarity matrices used

  2. a range normalization to bring the values back between 0 and 1 :

    $$\text{mat.DIST}_{i} = \frac{\text{mat.DIST}_{i} - min(\text{mat.DIST}_{i})}{max(\text{mat.DIST}_{i}) - min(\text{mat.DIST}_{i})}$$

See also

huge.npn

Author

Maya Guéguen

Examples


## Load example data
Champsaur_PFG = .loadData('Champsaur_PFG', 'RData')

## Species traits
tab.traits = Champsaur_PFG$sp.traits
tab.traits = tab.traits[, c('species', 'GROUP', 'MATURITY', 'LONGEVITY'
                            , 'HEIGHT', 'DISPERSAL', 'LIGHT', 'NITROGEN')]
str(tab.traits)

## Species niche overlap (dissimilarity distances)
DIST.overlap = Champsaur_PFG$mat.overlap
DIST.overlap[1:5, 1:5]

## Species functional distances (dissimilarity)
DIST.traits = PRE_FATE.speciesDistanceTraits(mat.traits = tab.traits
                                             , opt.maxPercent.NA = 0.05
                                             , opt.maxPercent.similarSpecies = 0.3
                                             , opt.min.sd = 0.3)
DIST.traits$Chamaephyte[1:5, 1:5]

## Combine distances ---------------------------------------------------------
list.DIST = list(DIST.overlap, DIST.traits$Chamaephyte)
sp.DIST.n = PRE_FATE.speciesDistanceCombine(list.mat.dist = list.DIST
                                            , opt.weights = c(0.2, 0.8))
sp.DIST.un = PRE_FATE.speciesDistanceCombine(list.mat.dist = list.DIST
                                             , opt.norm = FALSE
                                             , opt.weights = c(0.2, 0.8))
str(sp.DIST.n)



if (FALSE) {
require(corrplot)
list.DIST = list(DIST.overlap, DIST.traits$Chamaephyte
                 , sp.DIST.un, sp.DIST.n)
names(list.DIST) = c('overlap', 'traits', 'un-normed', 'normed')

par(mfrow = c(2, 2))
for (li in 1:length(list.DIST))
{
  tmp = list.DIST[[li]]
  tmp = tmp[colnames(sp.DIST.n), colnames(sp.DIST.n)]
  corrplot(tmp, method = 'shade'
           , type = 'lower', cl.lim = c(0, 1)
           , is.corr = FALSE, title = names(list.DIST)[li])
}

require(foreach); require(ggplot2); require(ggdendro)
hc = hclust(as.dist(sp.DIST.n))
pp = ggdendrogram(hc, rotate = TRUE) +
  labs(title = 'Hierarchical clustering based on species distances')
plot(pp)
}