Mantel test

ric

New Member
#1
Hi everyone,
I have a set of data where I analyse the effect of a 3 treatments on 3 different plants (the table is as follow):

plant 1 plant2 plant3
Sample 1 2 3 4 5... 1 2 3 4 5... 1 2 3 4 5...
variable1 a b c d e f g h i j k l m n o
V2
V3
V4
.
.
.
Sample 1-6 = treatment1; sample 7-12= treatmen 2 ;sample 13-18 =treatment 3
I did a PCA and cluster analysis but the results are messy, i can't see any clear separatibility between the sample belonging to the same treatment. But if I separate the 3 plants and I anayse the effect of the 3 treatment for each of the plants, the PCA and the cluster analysis gave me a very good separatibility and look very similar from one to an another. So I think the 3 plants are affected by the treatment the same way but because they are different from the beginning, the samples belonging to the same treatment and different plant don't cluster together. I have small clusters representing 1 plant and 1 treatment.
I was trying to use a method to minimise the variability between plants. I used mantel test to compare the similarity matrix a each plant and they appeared to be correlated. So i concluded that each plant is affected the same way by the treatments so I concatenate the 3 table of results (variable/samples) together:
Plant1: sample: 1 2 3 4 5 ....
Variable1: a b c d e ...
V2
V3
.
.
.
Plant2: Sample: 1 2 3 4 5 ....
V1
V2
V3
Plant3: Sample: 1 2 3 4 5 ....
V1
V2
V3
And then I did a PCA and cluster analysis on this concatenated table and the results are very good.
I don' t know if this is valid. I know Mantel is use to see correlation between for example a bacterial population and a site of sampling but when the samples are the same. In my case, the samples are similar (same time of sampling, same part of the plant, same processing) but belong to different plants.
Thanks for your help.
Is anyone has a better idea to decrease the variability between the plants or maybe a technique to create a consensus similarity matrix from the 3 I obtain with the 3 different plant.
ps: I tried Linear discriminant analysis it works but i dont't really like supervised techniques (even if they can be validated).