Performs principal components analysis (PCA) on a xyz numeric data matrix.

# S3 method for xyz
pca(xyz, subset = rep(TRUE, nrow(as.matrix(xyz))),
                 use.svd = FALSE, rm.gaps=FALSE, mass = NULL, ...)

# S3 method for pca
print(x, nmodes=6, ...)



numeric matrix of Cartesian coordinates with a row per structure.


an optional vector of numeric indices that selects a subset of rows (e.g. experimental structures vs molecular dynamics trajectory structures) from the full xyz matrix. Note: the full xyz is projected onto this subspace.


logical, if TRUE singular value decomposition (SVD) is called instead of eigenvalue decomposition.


logical, if TRUE gap positions (with missing coordinate data in any input structure) are removed before calculation. This is equivalent to removing NA cols from xyz.


an object of class pca, as obtained from function


numeric, number of modes to be printed.


a ‘pdb’ object or numeric vector of residue/atom masses. By default (mass=NULL), mass is ignored. If provided with a ‘pdb’ object, masses of all amino acids obtained from aa2mass are used.


additional arguments to (for or to print (for print.pca).


If mass is provided, mass weighted coordinates will be considered, and iteration of fitting onto the mean structure is performed internally. The extra fitting process is to remove external translation and rotation of the whole system. With this option, a direct comparison can be made between PCs from and vibrational modes from nma.pdb, with the fact that $$A=k_BTF^{-1}$$, where \(A\) is the variance-covariance matrix, \(F\) the Hessian matrix, \(k_B\) the Boltzmann's constant, and \(T\) the temperature.


Returns a list with the following components:




eigenvectors (i.e. the x, y, and z variable loadings).


scores of the supplied xyz on the pcs.


atom-wise loadings (i.e. xyz normalized eigenvectors).


the standard deviations of the pcs.


the means that were subtracted.


Grant, B.J. et al. (2006) Bioinformatics 22, 2695--2696.


Barry Grant

See also


if (FALSE) { #-- Read transducin alignment and structures aln <- read.fasta(system.file("examples/transducin.fa",package="bio3d")) pdbs <- read.fasta.pdb(aln) # Find core core <- core.find(pdbs, #write.pdbs = TRUE, verbose=TRUE) rm(list=c("pdbs", "core")) } #-- OR for demo purposes just read previously saved transducin data attach(transducin) # Previously fitted coordinates based on sub 1.0A^3 core. See core.find() function. xyz <- pdbs$xyz #-- Do PCA ignoring gap containing positions pc.xray <-, rm.gaps=TRUE)
#> NOTE: Removing 49 gap positions with missing coordinate data #> retaining 305 non-gap positions for analysis.
# Plot results (conformer plots & scree plot overview) plot(pc.xray, col=annotation[, "color"])
# Plot a single conformer plot of PC1 v PC2 plot(pc.xray, pc.axes=1:2, col=annotation[, "color"])
## Plot atom wise loadings plot.bio3d(pc.xray$au[,1], ylab="PC1 (A)")
# \donttest{ # PDB server connection required - testing excluded ## Plot loadings in relation to reference structure 1TAG pdb <- read.pdb("1tag")
#> Note: Accessing on-line PDB file
#> Warning: /var/folders/xf/qznxnpf91vb1wm4xwgnbt0xr0000gn/T//Rtmp9oBdbc/1tag.pdb exists. Skipping download
ind <- grep("1TAG", pdbs$id) ## location in alignment resno <- pdbs$resno[ind, !] ## non-gap residues tpdb <- trim.pdb(pdb, resno=resno) op <- par(no.readonly=TRUE) par(mfrow = c(3, 1), cex = 0.6, mar = c(3, 4, 1, 1)) plot.bio3d(pc.xray$au[,1], resno, ylab="PC1 (A)", sse=tpdb) plot.bio3d(pc.xray$au[,2], resno, ylab="PC2 (A)", sse=tpdb) plot.bio3d(pc.xray$au[,3], resno, ylab="PC3 (A)", sse=tpdb)
par(op) # } if (FALSE) { # Write PC trajectory resno = pdbs$resno[1, !] resid = aa123(pdbs$ali[1, !]) a <- mktrj.pca(pc.xray, pc=1, file="pc1.pdb", resno=resno, resid=resid ) b <- mktrj.pca(pc.xray, pc=2, file="pc2.pdb", resno=resno, resid=resid ) c <- mktrj.pca(pc.xray, pc=3, file="pc3.pdb", resno=resno, resid=resid ) } detach(transducin)