Splits and Networx

library(phangorn)
data(Laurasiatherian)
data(yeast)

consensusNet

A consensusNet (Holland et al. 2004) is a generalization of a consensus tree. Instead only representing splits with at least 50% in a bootstrap or MCMC sample one can use a lower threshold. However of important competing splits are left out.

The input for consensusNet is a list of trees i.e. an object of class multiPhylo.

set.seed(1)
bs <- bootstrap.phyDat(yeast, FUN = function(x)nj(dist.hamming(x)), 
    bs=100)
tree <- nj(dist.hamming(yeast))
par("mar" = rep(1, 4))
tree <- plotBS(tree, bs, "phylogram")

cnet <- consensusNet(bs, .3)
plot(cnet, "2D", show.edge.label=TRUE)

Often consensusNet will return incompatible splits, which cannot plotted as a planar graph. A nice way to get still a good impression of the network is to plot it in 3 dimensions.

plot(cnet)
# rotate 3d plot
play3d(spin3d(axis=c(0,1,0), rpm=6), duration=10)
# create animated gif file 
movie3d(spin3d(axis=c(0,1,0), rpm=6), duration=10)

which will result in a spinning graph similar to this

neighborNet

The function neighborNet implements the popular method of Bryant and Moulton (2004). The Neighbor-Net algorithm extends the Neighbor joining allowing again algorithm is computed in 2 parts, the first computes a circular ordering. The second step involves estimation of edge weights using non-negative Least-Squares (nnls).

dm <- dist.hamming(yeast)
nnet <- neighborNet(dm)
par("mar" = rep(1, 4))
plot(nnet, "2D")

The advantage of Neighbor-Net is that it returns a circular split system which can be always displayed in a planar (2D) graph. The plots displayed in phangorn may not planar, but re-plotting may gives you a planar graph. This unwanted behavior will be improved in future version. The rendering of the networx is done using the the fantastic igraph package (Csardi and Nepusz 2006).

Adding support values

We can use the generic function addConfidences to add support values from a tree, i.e. an object of class phylo to a networx, splits or phylo object. The Neighbor-Net object we computed above contains no support values. We can add the support values fro the tree we computed to the splits these two objects share.

nnet <- addConfidences(nnet, tree)
par("mar" = rep(1, 4))
plot(nnet, "2D", show.edge.label=TRUE)

We can also add support values to a tree:

tree2 <- rNNI(tree, 2)
tree2 <- addConfidences(tree2, tree)
# several support values are missing
par("mar" = rep(1, 4))
plot(tree2, show.node.label=TRUE)

Estimating edge weights (nnls)

Consensus networks on the other hand have information about support values corresponding to a split, but are generally without edge weights. Given a distance matrix we can estimate edge weights using non-negative Least-Squares.

cnet <- nnls.networx(cnet, dm)
par("mar" = rep(1, 4))
plot(cnet, "2D", show.edge.label=TRUE)