• Jean-Pierre Dujardin, IRD, Montpellier (France)

Features of “PAD” (“Permutaciones, Analisis Discriminante”)

  • Performs Discriminant analyses with limited graphical display (button “Mapa factorial“)
    • Computation is based on the inverse of the consensus covariance matrix
    • Factor map of the first two discriminant factors (DF1,DF2) only
    • Respective contributions of each discriminant factor
    • Table of Mahalanobis distances, formatted as for PHYLIP (J. Felsenstein) tree construction.
  • Permutation of individuals among groups (button “Permutar distancias ?“)
    • Random permutation of individuals in the total sample is used as a method to randomly interchange individuals among groups
    • At each iteration, a complete discriminant analysis is performed
    • As an estimate of statistical significance for each Mahalanobis distance, a proportion is given which is the number of Mahalanobis distances values equal or higher than observed, relative to the total number of permutations.
  • Size-free analysis (button “DISTANCES, or TRADITIONAL“)
    • Input data MUST be log-transformed measurements
    • Size-free variables are computed (row centered data)
    • Principal components are derived from these size-free variables, called here “size-free components”
    • The optimal number of size-free components is suggested in relation to local sample sizes
    • A discriminant analysis is performed on this number of size-free components.
    • Allometric residual may be examined on each of the two first discriminant factors.
  • Coordinates of landmarks analysis (button “LANDMARKS, or GEOMETRIC“)
    • Input data are always Partial Warps, they are used either without restriction (button “PW, option ‘Total’) or “partly” (button “PW, option ‘Part’), meaning that only a few first Relative Warps are used. Select that option when the number of PW is too high relative to the number of individuals in the smallest group.
    • Thus, if the option “PW, option ‘Part’ is selected, Relative Warps are computed and a subset of them is suggested for subsequent analyses.
    • Allometric residual may be examined on each of the two first discriminant factors, provided a file is opened containing corresponding ‘centroid sizes’ (CS).
  • Supplementary data (button “Datos Externos“)
    • External data may be called, contained in a FORMAT formatted file.
    • Their position on the DF1/DF2 factor map is computed according to the first two discriminant factors, DF1 and DF2
    • In case of two groups, supplementary data are visible on the histogram of discriminating classes(March 2009)
    • Mahalanobis distances are computed between each external individual and each centroid of the discriminant model
    • A classification table is given
    • This procedure is NOT available anymore for PW data; for a shape based Mahalanobis classification of unknown individuals relative to 2 or more groups, please use the MOG module.
  • Reclassification of individuals (button “C“)
    • This corresponds to the same sample being treated as external data
  • Cross checked reclassification (button “CCC“).
    • For each individual classification the individual itself is removed from the total sample, then:
      – sample sizes are computed again;
      – the Mahalanobis distances are computed between new centroids;
      – the removed individual is then used as supplementary data, and its distance computed with each centroid;
      – a re-classification table is contructed.
  • External softwareis available (menu)
    • The PHYLIP ‘neighbor’ and ‘fitch’ modules (J. Felsenstein) may be called directly from PAD; people wanting to use other modules of the PHYLIP package should go here.
    • After answering the questions by the letters ‘N‘ (to obtain the UPGMA method instead of the Neighbor-Joigning one), ‘L‘ (lower diagoal matrix) and ‘Y‘ (yes) …
    • … the PHYLIP outfile is directly printed within the PAD report.
    • Each OTU is labeled as a number according to its order position in the input file. Giving a name to each OTU is possible by the menu “External Software” -> “Giving a name to each OTU”
    • Direct access is also given to an editing tree software, njplot (see: Perriere, G. and Gouy, M. (1996) WWW-Query: An on-line retrieval system for biological sequence banks. Biochimie, 78, 364-369).



  • CLIC package


  • Initial panel
  •   Size contribution to discrimination
  •   Three groups
  • Two groups and external data