[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[modeller_usage] PCA problem



Dear all,


My question deals with the command which performs Principal Component Analysis (PCA). I use an input matrix of pairwise distances

calculated from percentage residue identity of a multiple protein sequence alignment.

I executed the following script :


#################

env = environ()

aln = alignment(env, file='alignment.pir')

aln.id_table(matrix_file='id.mat')

env.principal_components(matrix_file='id.mat', file='alignment_mod..princ')

#################


The file 'alignment_mod.princ' contains projected coordinates of the first two principal components..

I compared these results with other programs which hold PCA and go by the same input matrix.

For instance, ade4, a package of the R software.


I executed the following commands under R :


#################

require (ade4)

x <- read.table("id.mat", sep="")

y <- dudi.pca(x, scan = F, nf = 2)

write.table(y$li, file = "alignment_ade4.princ")

#################


The file 'alignment_ade4.princ' also contains projected coordinates of the first two principal components.


However, the coordinates acquired with these two softwares are not the same.


I would like to know if the PCA command of Modeller uses internal procedures which could explain this disagreement ? How does the treatment of input matrix work with Modeller ?


Any help on this will be greatly appreciated.


Best regards, Julien