Block.splsda - DIABLO - correlation between samples and variables

Analysis
1

Hi,

Is there any way to extract the association between samples (Y) and selected variables (biomarkers) ? like what we get in case of splsda?

Thanks a lot and looking forward

2

hi @nomranian,

It depends on the method you use. We have created a biplot for PCA and PLS.
Otherwise, we usually inspect the correlation circle plot in parallel with the sample plot to work out the associations.

Kim-Anh

3

Hi @kim-anh
Thanks a lot for your response.
I did the DIABLO analysis and I get as a network only association between the entities but not between entities and samples like what I could get from splsda.

I appreciate a lot your help.

4

hi @nomranian,

I am not sure which output you are referring to for DIABLO to extract the association between samples and variables. As I said earlier, you can examine the correlation circle plots AND the sample plots co-jointly to infer some associations. You can have a look at this vignette for a basic example on PCA and apply the same principles for the other methods (section 3.2): 3 PCA on the multidrug study | mixOmics vignette

You can also have a look at the example for N-integration and see if there is any output that you fancy.

Kim-Anh

5

Hi @kimanh.lecao
Thank you very much for your response.
I can tell you what I mean wrt the example given here:
http://mixomics.org/mixdiablo/diablo-tcga-case-study/
three blocks (miRNA, mRNA , proteomics) and three conditions (Basal , Her2, LumA)

The network plot:
network(final.diablo.model, blocks = c(1,2,3),
color.node = c(‘darkorchid’, ‘brown1’, ‘lightgreen’), cutoff = 0.4)

only presents association between the key elements of each block.
What I’m looking for, as I can also get from splsda, the same network but between elements and conditions. I want to know how discriminative biomarkers are correlated to the conditions.

Is this possible with block.splsda (DIABLO)?

Thanks and looking very much forward.

6

hi @nomranian,

Ah, I see. No we have not implemented this version of the network. Beside considering both correlation circle plots + sample plot in your interpretation, you can consider using plotLoadings. Another plot of interest could be the circosPlot. The lines on the outside of the plot show the mean expression values per group (see here: circos() | mixOmics)

Kim-Anh

7

Hi @kimanh.lecao
Thanks a lot for your response.
Actually what I want is not the plot but the correlation strength values between the elements and conditions (between miRNA, mRNA or proteins identified as biomarkers and conditions such as Basal, Her2 and LumA)
But if this is not implemented then I will try to see how I can extract it from the code because it should not be difficult.
Thanks again :blush:

8

hi @nomranian,

Yes for the correlations you can extract them from cimDiablo, and circosPlot if you save them as an object. Similar for plotLoading if you need extra information.

Kim-Anh