Understanding interpretation of higher percent variance in Component 2

1

Hello! I am very new to PLS-DA/SPLS-DA and would like to ask about interpretation of my results.

I made a SPLS-DA with some metabolomic data to determine some biomarkers between two treatment groups. The separation of the groups separated nicely, although I am having a little bit of understanding on how to best interpret the data based on the variance/components.
I have usually done more work with PCA/PCoA and so understand that the components are always ordered by variance, so higher variance found in component 1 than 2.

However, when I do my SPLS-DA, since the components are based more on discriminating versus variance, I found my component 2 had a higher variance % compared to my component 1.

When interpreting this data, what is the best way to go about this? If it isn’t variance that is most driving the discrimination between the two groups what is?

2

hi @lilyyycao,

Welcome!
Yes, there is a clear distinction between PLSDA and PCA.
In PCA you want to maximise the variance explained by each component, as the variance tells us how much information we can summarise from the data.

In PLSDA we maximise the discrimination between sample groups, and this may not correspond to the highest variance (if it does, it’s a great sign that your groups can be separated, but it if does not, it tells you that the major source of variation may be due to something else in the data). So in PLSDA the amount of explain variance is just additional information about your data and how separable your samples are, but it is not an important result. In PCA we report the amount of explained variance, but in PLSDA we are mostly interested in the classification performance (e.g perf()).

Kim-Anh