[R-sig-eco] betapart and nms

[Mitchell, Kendra]

I'm working on a large bacterial dataset, 700 samples with ~5600 species (rare species <100 observations have been removed).  The samples are from experimental plots from 6 ecozones.  I've done nms, betadisp, and permanova using several dissimilarity measures (jaccard, sorenson, Yuen's theta).  Ecozone explains the vast majority of the difference between the samples using those measures.  I'm really interested in trying to find treatment effects so have also tried betapart-specifically trying to see if nestedness increases in the harsher treatments.  

Here is where I need input from the list.  beta.sim gives a similar ordination to the dissimilarity measures and the expected step down in stress as I increase number of dimensions. However beta.sne is very strange and I don't know how to interpret it.  The stress in 1d is >0.6, but drops to 0 in 2d.  When I plot the 2d solution I get a horseshoe-it looks like the typical distortion in a PCA of nonlinear data.  I've run a lot of NMS on these types of nonnormal, empty, huge matrixes and have never seen this sort of pattern in the ordination and my stress never goes much below .15.  I'm not sure how to assess what beta.sne is calculating to help me make sense of this result.  I've subsampled my species matrix a few times just to make sure something strange didn't happen with the subsampling.

thanks for reading, Kendra




Here's what I've run so far

##betapart

#subsample matrix in mother to 1950 observations per sample

b03_f100_1_1 <- read.table("bac_03_f100_sub1", row.names=1, header=T)

#convert to presence/abscence
b03_f100_1_pa <- decostand(b03_f100_1_1, method="pa")


#betapart
bac03_bp<-beta.pair(b03_f100_1_pa, index.family="sorensen" )

#nms on betapart objects
scree_bsim1<-nmds(bac03_bp$beta.sim, mindim=1, maxdim=6, nits=10)      #stress: 1d .65, 2d .35, 3d .25, 4d .19
scree_bsne1<-nmds(bac03_bp$beta.sne, mindim=1, maxdim=6, nits=10)      #stress: 1d .6, 2d 0, 3d 0...


bac_bsim_nms<- metaMDS(bac03_bp$beta.sim, k=3, trymin=50, trymax=250, wascores=FALSE)
bac_bsne_nms<- metaMDS(bac03_bp$beta.sne, k=2, trymin=50, trymax=250, wascores=FALSE)


--
Kendra Maas Mitchell, Ph.D.
Post Doctoral Research Fellow
University of British Columbia