I have performed rCCA between transcriptomics and proteomics datasets. But in the correlation matrix i got values greater than 1. What is the meaning of correlation greater than 1. The commands I executed are pasted below.
Corr_Shrink <- rcc(Prot, Rna, ncomp = 3, method = ‘shrinkage’)
cluster=cim(Corr_Shrink, comp = 1:3,margins = c(5, 6),threshold = 0.8)
write.table(as.data.frame(cluster$mat.cor),file=“Correlation.txt”,sep="\t",quote = FALSE)
Hi Anoop,
The cim actually adds the correlations from one component to the other. It seems that 3 components is ‘over’ 1. Specify either component 1 or component 1,2 in your cim and let me know how you go (ps: I could not see your image).
I had drafted a long answer for the case where plotVar() variables would be outside the radius of 1. If it is the case let me know and I’ll include my other answer!
Also report the canonical correlation from your rcc object, it should be < 1 for each component.
Kim-Anh
HI Kim,
I tried cim with comp=1, but still getting correlations greater than 1.
ec_cluster=cim(ec_Corr_Shrink,margins = c(5, 6),threshold = 0.8,row.names = FALSE,col.names = FALSE,comp = 1)
Hi @Anoop.
Would you mind sharing your data + reproducible code at mixomics[@]math.univ-toulouse.fr? We can then assess whether it is coming from the data, or a potential limitation of CIM. As far as I know both plotVar() and CIM would use the same similarity matrix.
In the meantime you can focus your interpretation on the correlation circle plot (details in https://biodatamining.biomedcentral.com/articles/10.1186/1756-0381-5-19)
Kim-Anh
Hi,
W can share the data. But I am not able to find where to upload the data in the link you have given. Could you please help in that
Please send us an email by replacing [@] with a normal @.
Hello everyone,
I was wondering whether you had any updates with regards of the issue experienced with cor values being >1 for a single component. Was that due to data or the method itself?
Thanks a lot in advance
bests
Joosè
Hi @joosefupas,
This issue was in fact fixed, but as discussed in this topic, the cimDiablo function depicts the values - which are potentially unbounded - and not the correlations.
Hope it helps,
Al
