sPLS-DA:srbct

Samples representation

Let us have a first glimpse at the individual plots for the first 2 principal components

plotIndiv(result, comp = 1:2, ind.names = FALSE, col = col.srbct, cex = 1, pch = 16)

## Add a legend
col.legend <- unique(col.srbct)
legend(0.12, -0.11, c("Ewing's sarcoma", "Burkitt's lymphoma", "Neuroblastoma",
       "Rhabdomyo sarcom"), col = col.legend, pt.cex = 1, pch = 16,
       title = "Tumour Class")

The clusters of the classes are much better. In 3D:

plot3dIndiv(result, comp = 1:3, col = col.srbct, cex = 0.25, axes.box = "both")

Using PLS-DA with no variable selection

This is what happens if instead we used a classical PLS-DA without variable selection:

## PLS-DA with no variable selection
result.pls <- splsda(X, Y, ncomp = 3)
plotIndiv(result.pls, comp = 1:2, ind.names = FALSE, col = col.srbct,
          cex = 1, pch = 16)

We still include information about the classes of the samples, but many of the genes are noisy and uninformative regarding the class of the samples. This is why without variable selection the clusters are less defined than in the sPLS-DA case.

Next: Variables representation

Pages: 1 2 3 4