Doutoranda em ecologia, laboratorio de abelhas, orientada por Isabel Alves dos Santos. Tema de projeto de tese: fluxo de polen entre fragmentos e áreas restauradas de Mata Atlântica.

exec

PROPOSTA 1

Fazer uma função com ciclos de regressões lineares simples para identificar es variáveis preditoras de uma variável 'y' de vários níveis (ou classificada por um fator). O produto da função é uma tabela com o resultados das regressões (R2 e valor de p) entre cada nível da variável dependente e as variáveis preditoras.

Exemplo:

Pergunta: Qual o quais são as variáveis que melhor explicam a abundância das espécies em diferentes locais (fragmentos de vegetação).

• Variável dependente: abundancia de cada espécie.
• Variáveis preditoras: associadas as características dos locais (área do fragmento, forma do fragmento, grau de isolamento, etc.)

Resultado:

PROPOSTA 2

Uma função para fazer uma regressão entre os coeficientes de similaridade de pares de amostras (i.e. variável dependente) e as diferencias entre os valores das variáveis preditoras dos pares de amostras. A função calculará os coeficientes de similaridade e as diferencias entre pares de amostras e depois fará a regressão.

Exemplo: O mesmo anterior.

1. calcular matrizes de coeficientes de similaridade segundo o numero de espécies coexistentes entre locais

2. calcular matrizes de diferencias entre os valores das variáveis preditoras e logo fazer o seguinte data.frame:

3. Fazer uma regressão com os dados do data.frame

require(vegan)

####Input two data sets

sophisticated.mantel <- function(datasp, datavar, sample.dist ="bray", var.dist="euclidean", cor.method ="pearson", ...) 
{
  if(is.numeric(datasp) | is.numeric(datavar))
    stop("data must be a data frame or a matrix") #data set class must be data frame or matrix
  x <- vegdist(datasp, method = sample.dist, ...) #sample dissimilarity matrix

##Tables of results   
  result <- data.frame(Variables = colnames(datavar), Mantel_statistic_r = rep(NA, ncol(datavar)), Significance = rep(NA, ncol(datavar))) #Table of Mantel test results
  result.sp <- matrix(NA, ncol(datasp), ncol(datavar)+3) #Table of Species analysis results
  colnames(result.sp) <- c("Species", "Abundance", "Frecuency", colnames(datavar))
  result.sp[,1] <- colnames(datasp) #Species names in the tables's first column
  
##Mantel test for each independent variable  
  for(i in 1:ncol(datavar))
  {
    y <- dist(datavar[,i], method = var.dist, ...) #distance matrix for each independent variable
    z <- mantel(x,y, method = cor.method) #Mantel test between distance matrixes of samples and independent variables                                                                        
    result$Mantel_statistic_r[i] <- round(z$statistic, 3) #R2-values from the Mantel test
    if(z$signif <= 0.05){
      result$Significance[i] <- paste(z$signif, "*") #significant p-values from Mantel test 
    }
    else{
      result$Significance[i] <- paste(z$signif, " ") #non-significant p-values
    }  
  }

#Species analysis
  for(k in 1:ncol(datasp))
    {
      result.sp[k,2] <- sum(datasp[,k]) #total abundance of each species
      result.sp[k,3] <- sum(datasp[,k]!= 0) #total frecuency of each species
      for(b in 1:ncol(datavar)){
        d <- anova(lm(datasp[,k]~1),lm(datasp[,k]~datavar[,b])) #linnear regression between each species abundances and each independent variable
        p <- round(d[2,6], 3) #p-values
        if(d[2,6] <= 0.05){
          result.sp[k,b+3]<- paste(p,"*") #significant p-values  
        }
        else{
          result.sp[k,b+3] <- paste(p, " ") #non-significant p-values
        }
      }
    }
  return(list(Mantel_Test_Results = result, Species_Analysis = data.frame(result.sp)))   #list of results
 }
   
##Examples

data(varespec)
data(varechem)

sophisticated.mantel(varespec, varechem)

sophisticated.mantel             package:unknown                R 

Documentation

A sophisticated Mantel test for species community analysis
  
  
  Description:
  
  This function creates distance matrices from a first data set of species abundances in samples and a second data set of explanatory variables from samples. Afterwards runs the mantel test for each explanatory variable. In addition, the function runs simple linear regressions between each species and each explanatory variable.

  
  Usage:
  
  sophisticated.mantel(datasp, datavar, sample.dist ="bray", var.dist="euclidean", cor.method ="pearson", ...)
  
  
  Arguments:
  
  datasp:       a data frame or matrix of species abundances (columns) distributed in samples (rows).

  datavar:      a data frame or matrix of numeric independent variable(s) associated to samples. Row length must be the same as datasp.

  sample.dist:  the distance measure to be used for computing the sample/species distance matrix (from datasp) using 'vegdist'(from vegan package). This must be one of "manhattan", "euclidean", "canberra", "bray"(default), "kulczynski", "jaccard", "gower", "altGower", "morisita", "horn", "mountford", "raup", "binomial", "chao" or "cao". 
  
  var.dist:     the distance measure to be used for computing distance matrices of the explanatory variables (from datavar), using 'dist'. This must be one of "euclidean" (default), "maximum", "manhattan", "canberra", "binary" or "minkowski".

  cor.method:   the correlation coefficient to be computed by 'mantel' (from vegan package) as accepted by 'cor'. This can be "pearson"(default), "spearman" or "kendall".
  
  ...           additional arguments from the functions 'vegdist', 'dist' and 'mantel'. 
  
  
  Details:
  
  'datasp' is converted to a distance matrix which is calculated with the distances between samples. 
  
  'datavar' is converted to one or many distance matrices according to the number of variables (columns). Each matrix is calculated with the differences (distances) between the variable's values on every pair of samples. Factor variables must be arranged as binary data on 'datavar', with as many columns as the number of factor levels.

  The function returns the coefficients of the correlations between the distance matrix of each variable and the distance matrix of the sample/species matrix. Significance values are calculated by permutation analysis (see 'mantel' in vegan package documentation).

  The second part of the function returns a data frame with the species's total abundances and frecuencies, as well as the significance values of simple linear regressions between species abundances and explanatory variables using 'lm'.  

 
  Value:
  
  The function returns a list with the following components:

  Mantel_Test_Results:  R-square values ("Mantel_statistic_r") and p-values ("Significance") of the correlation with each explanatory variable.

  Species_Analysis:     p-values of simple linear regressions between species abundances and explanatory variables. The asterisk '*' indicates statistically significant relations.

  
  Warning:
  
  Data sets must be data frames or matrices with the same row length and must not have NA values. 
  

  Note:
  
  Legendre and Legendre (2012) explain distance methods and give suggestions for using them depending on data characteristics. Distance method explanations are also given at 'vegdist' vegan package documentation. 
  
  
  Author(s):
  
  Paula Maria Montoya-Pfeiffer
  
  
  References:
  
  Legendre, P. and Legendre, L. (2012) Numerical Ecology. 3rd English Edition. Elsevier.
  
  Jari Oksanen, F. Guillaume Blanchet, Roeland Kindt, Pierre Legendre,
Peter R. Minchin, R. B. O'Hara, Gavin L. Simpson, Peter Solymos,
M. Henry H. Stevens, Helene Wagner (2015). Community Ecology Package 'Vegan'. 280pp. Available in: http://cran.r-project.org, https://github.com/vegandevs/vegan.
  
  
  See Also:
  
  'vegdist' and 'mantel' in vegan package and 'dist' for additional arguments and further information.
  
   
  Examples:
  
  require(vegan)
  data(varespec)
  data(varechem)
  
  sophisticated.mantel(varespec, varechem) ##with default arguments and data frames
  
  matrix.sp <- as.matrix(varespec) 
  sophisticated.mantel(matrix.sp, varechem) ##with a matrix data set
  
  sophisticated.mantel(varespec, varechem, sample.dist ="morisita", var.dist="canberra", cor.method ="spearman") ##with different distance methods

sophisticated_mantel.r