I have performed sPLS-DA on a gene expression dataset with 5 sample classes. I got a 4 component model with number of variables selected along the 4 components being 357,357,357 and 179 respectively. The sample plots, loading plots and clustered image maps have been attached here.
I wish to find features that can discriminate ‘TNBC’ samples from others. Similarly, I wish to find features that can discriminate ‘Normal’ samples from others. As these classes are not getting segregated well in one particular component, I am finding it difficult to interpret the result. Any help is highly appreciated.
Thanks!
Sample space on component 1 and 2:
Sample space on component 2 and 3:
Loading plot on component 1 with contrib = “max”:
Loading plot on component 1 with contrib = “min”:
Loading plot on component 2 with contrib = “max”:
Loading plot on component 2 with contrib = “min”:
Clustered image map on component 1:
Clustered image map on component 2:
hi @srikantverma
You can save the plotLoadings() into an object and from there extract the column GroupContrib
data(liver.toxicity)
X = as.matrix(liver.toxicity$gene)
Y = as.factor(liver.toxicity$treatment[, 4])
splsda.liver = splsda(X, Y, ncomp = 2, keepX = c(20, 20))
plot.contrib = plotLoadings(splsda.liver, comp = 2, method = ‘median’, plot = FALSE,contrib = “max”)
plot.contrib
You can do similar for the CIM to save/extract the matrix.
Kim-Anh
Thanks a lot @kimanh.lecao for your kind help!
This will certainly help to extract features relevant to a particular class of samples on a particular component. My next question is how to decide which component to choose for a particular class, say TNBC in this case. The PlotIndiv plots show that component 1 discriminates both TNBC and Normal samples from others, while component 2 further discriminates TNBC from Normal. So, should I consider both component 1 and 2 for extracting TNBC relevant features?
Your input means a lot to me, and other mixomics users.
Thanks & regards
Srikant
