dotinfer: Inference of a Sympatry Network from Point Data
In SyNet: Inference and Analysis of Sympatry Networks

Description Usage Arguments Details Value Note Author(s) References See Also Examples

Using a Graphical User Interface (GUI), this function explores the pairwise relationships scored by weighted matrices of spatial association. It also helps to choose the cutoff value, creates an interactive network and plots intensity matrices.

1	dotinfer(dotdata, mtxdata)

`dotdata`	Object of class 'dotdata'.
`mtxdata`	List of weighted matrices based on the previous data. The names for each element of this list are considered as the identity labels of matrices.

The list of weighted matrices in addition to the list of involved taxa are loaded into two different combo boxes. It is very important to identify the right nature of the data (i.e. dissimilarity/similarity). In similarity matrices there is a direct relationship between their scores and the strength of association. On the contrary, in dissimilarity matrices there is an inverse relationship between their scores and the strength of association.

Once you have selected a given matrix of spatial association, its single linkage dendrogram is drawn. On the other hand, after you choose a given focus species, its couples are decreasingly ordered by the strength of spatial association between them. The upper left panel shows the spatial distributions of species pairs under consideration. The current focus species is indicated through red dots, whereas the current couple is represented by blue dots. There are buttons to shift the couples of the focus species, thus the user can move from more to less associated neighbors according to the values dictated by the matrix under analysis.

Thresholds are used to dichotomize a weighted matrix into a binary one. I know this procedure entails loss of information. In order to preserve the largest amount of information, the bottom right panel assists the user in this task. Thus, the strongest relationships found on the matrix are recovered using the notion of stable couples (Gale and Shapley 1962). The stable matchings are assumed to be representative of the strongest links in the underlying weighted network. Then, the distribution of affinity scores across the stable couples is estimated via kernel density estimates. The percentage under the density curve can be used as objective guideline to select the cutoff value.

Two appealing features of this interface can be found at the Display menu : i) manipulations of VAT images and ii) interactive layout of the inferred sympatry network.

The VAT approach presents pairwise association information about the set of objects O = {o_1, ..., o_n} as a square digital image with n*n pixels, after the objects are suitably reordered so that the image is better able to highlight potential cluster structure. VAT operates on the matrix of association between items and transforms each score into a value in the gray tone scale.

The network derived from thresholding is displayed in an interactive graph drawing facility. Here, nodes can be picked out and moved throughout the canvas (in addition to its incident edges). The right click on a given node highlights the target node and its neighbors in red and blue colors, respectively.

An object of class 'dotinference' is created into the working environment. It is a list with components:

`sm`	An adjacency matrix that suggests (1) or not (0) the occurrence of a sympatric link between species.
`Label`	Character vector of species labels.
`occupancy`	List of records occupied by each species.
`coords`	Spatial coordinates for each record occurring in the data set. They are arranged into a two columns matrix.
`kind`	Character. Specifies the kind of distributional data, that is "points".

Intensity matrices are produced with the VAT algorithm published by Bezdek and Hathaway (2002). Certainly, improvements on the layout of the network need to be introduced. I hope to work on that in the near future!

Daniel A. Dos Santos, <dadossantos@csnat.unt.edu.ar>

Bezdek J.C., Hathaway R. 2002. VAT: A Tool for Visual Assessment of (Cluster) Tendency. Proc. Int. Joint Conf. Neural Networks (IJCNN 2002): 2225-2230.

Dos Santos D.A., Cuezzo M.G., Dominguez E., Reynaga M.C. 2011. Towards a Dynamic Analysis of Weighted Networks in Biogeography. Systematic Biology (in press).

Gale D., Shapley L. 1962. College Admissions and the Stability of Marriage. Amer. Math. Month. 69:9-15.

Objects of class 'dotdata' result from submitting raw punctual data to the function procdnpoint. Kernel density estimates are calculated with the density function. Stables couples are obtained with the standard arguments available at stablecouple.

  data(mayflynz)
  aux <- procdnpoint(mayflynz) #Pre-processing of data
  toponz <- toposimilar(aux) #Similarity matrix. 
  acshnz <- acsh(aux) #Dissimilarity matrix. 
  rewnz <- reweight(toponz) #Similarity matrix. 
  #Explore the content of previous matrices in addition to the distributions
  #of involved species. 
  ## Not run: 
  dotinfer(aux, list(toponz = toponz, acshnz = acshnz, rewnz = rewnz))
  
## End(Not run)