1

Here I have another 'graphical' problem:

I have obtained from MOTHUR the following distance matrix (coming from a weighted unifrac analysis):

20
F3D0      
F3D1        0.222664
F3D141      0.157368    0.293308
F3D142      0.180278    0.319198    0.0944511
F3D143      0.157659    0.290975    0.0545202   0.0761392
F3D144      0.199909    0.34045 0.104358    0.086418    0.089473
F3D145      0.207946    0.348532    0.107841    0.076302    0.0940067   0.051632
F3D146      0.117877    0.253996    0.0891617   0.130867    0.0882064   0.134407    0.138415
F3D147      0.197256    0.336583    0.102114    0.0764106   0.0890669   0.0514887   0.0479297   0.135324
F3D148      0.173824    0.311951    0.0606815   0.0648557   0.056463    0.074914    0.0811015   0.111996    0.0709027
F3D149      0.145614    0.276632    0.0462779   0.105512    0.0628737   0.10902 0.114584    0.0739466   0.107123    0.0690412
F3D150      0.129557    0.277624    0.0840909   0.128305    0.0863231   0.140256    0.145381    0.0744572   0.13672 0.113564    0.0659831
F3D2        0.133531    0.216587    0.160832    0.186833    0.176061    0.214934    0.215261    0.152591    0.205629    0.188325    0.156313    0.153841
F3D3        0.213102    0.305651    0.123818    0.113021    0.139376    0.148558    0.13853 0.174377    0.139851    0.126329    0.131294    0.166738    0.137784
F3D5        0.128668    0.185235    0.167733    0.205183    0.176585    0.224806    0.230984    0.14497 0.223492    0.18933 0.153624    0.148617    0.127574    0.192433
F3D6        0.139411    0.236633    0.135418    0.124848    0.134198    0.175098    0.166205    0.118905    0.166144    0.151842    0.120964    0.12724 0.0950943   0.119852    0.129523
F3D7        0.198884    0.315888    0.130385    0.0989168   0.131945    0.14625 0.126203    0.173689    0.128993    0.121373    0.140199    0.152123    0.152893    0.0906675   0.186674    0.111134
F3D8        0.178656    0.18783 0.205737    0.22104 0.219858    0.268701    0.2644  0.184943    0.268051    0.229503    0.1979  0.20035 0.164427    0.203089    0.119084    0.142398    0.185551
F3D9        0.153265    0.186706    0.196143    0.21504 0.20728 0.262127    0.255558    0.174563    0.2607  0.221969    0.192437    0.185154    0.13976 0.195538    0.0973901   0.127619    0.177605    0.0558726
Mock        0.653789    0.645344    0.633297    0.623553    0.633903    0.633135    0.63394 0.635815    0.645332    0.636453    0.629143    0.646918    0.663222    0.639517    0.649722    0.64073 0.654882    0.63988 0.646155

As this distance matrix come from a PCoA, what I want to do is to plot these distances in an ordination plot with R.

Any idea on how to doing this?

Thanks a lot

www
  • 38,575
  • 12
  • 48
  • 84
mafernandez
  • 23
  • 5

2 Answers2

0

You have the vegan library with metaMDS function that generates coordinates for each sample using such a distance matrix as the input.

Let's call M to your matrix, you need to run this code:

    # Load the library
      library(vegan)
    # Use metaMDS function for 2D - plot
      NMDS <- metaMDS(distance = M, k = 2)
    # Plot your individuals
      plot(NMDS$points[,1], NMDS$points[,2])

In NMDS$points you have the coordinates for each of the samples. I suggest to colour the individuals according to a factor of interest such as cases and controls for example in biomedical analyses.

R18
  • 1,476
  • 1
  • 8
  • 17
  • Thanks a lot. Anyway, I have a doubt: 'metaMDS' does not perform an additional distance analysis to my data? Since these data come from another 'distance analysis', the only thing I need to do is to get the coordinates for each point (named after the row names) and plot them. – mafernandez Jul 06 '17 at 11:02
  • It just finds the best coordinates for the samples, such as the rank of given distances between individuals is conserved as much as possible. You can define the distance, but I you already have the distance matrix, just define it as `distance` parameter. – R18 Jul 06 '17 at 12:11
  • Hi again! Thanks a lot for your interest in my issue, @R18. I have tried your code, but with it directly I get a message telling that I need a 'comm' element to run metaMDS; As I have seen, 'comm' elements are community table data. I have also seen that a 'dist' element can be used, so I ran the code like: " NMDS <- metaMDS(M, distance = M, k = 2)" and it worked. However, the result is pretty weird, with all points except one in almost the same position. Is anything wrong with this trial? – mafernandez Jul 06 '17 at 14:04
  • `comm` should be your raw data; but as you have computed before the distance matrix, that is why the second option (the one I suggested) works. You have one sample far from the others. Let me guess its name . . . `Mock`? Just see the distance matrix and you can view that it is far from the rest of the individuals with higher distance values. – R18 Jul 06 '17 at 14:17
  • Thanks a lot. As you realized, the 'Mock' community was really different from the rest. I tried to run the commands on a file that has not the 'Mock' line and the plot was really informative. I will now post an answer with the code I have finnally used to do this. – mafernandez Jul 07 '17 at 10:09
0

Thanks to @R18, finally I could manage with this issue.

For the distance table I uploaded, the solution that I reached was to use the following code:

library(phyloseq)
library(vegan)
M <- import_mothur_dist("pcoa_UFdistance_matrix.dist")
unifrac <- metaMDS(M, distance = M, k = 2, trymax=100)
plot(unifrac$points[,1], unifrac$points[,2], main="Principal Coordinates Analysis", col.main="red", font.main=4, xlab="PCoA 1", ylab="PCoA 2")
text(unifrac, pos=3)

Hope it will help someone!!

mafernandez
  • 23
  • 5