is there a method like PLS-DA, but for a quantitative Y value?

Question

[boaty]

Hi guys,

I'm using PLS-DA to select the most importance genes contributing to group differences. Now, I am looking for a statistical method that fit quantitative Y.

For a normal PLS-DA, we have X (n*p) matrix which contains quantity data and Y (n) vector with qualitative information such as treatment/control or difference group. So PLS-DA will give us an optimal number of components and variables which maximize the separation between matrix and qualitative groups.

But what if I replace the Y value by quantitative value like fold-change? I know PLS which searches the best choice of components, but it is between 2 matrices. So for quantitative matrix vs quantitative vector, is there any statistical tools already existing?

Thanks in advance

Answer 1

[Kevin Blighe]

I'm not sure but perhaps my own Bioconductor package, PCAtools, addresses this issue. Take a look at sections 4.4 and 4.5 of the vignette:

• PCAtools: everything Principal Component Analysis

Kevin

[boaty]

thanks Kevin for your answer,

It is exactly what I am looking for! In your package, you are using loadings to classify and calculate correlations between PCs and category names(eg. Size, Grade etc.). In my case, only 1 category : log2(fold-change). I do not know if log2 ratio is linear or maybe not, but I can use rank test anyway.

This will give us the information about which PC correlates with fold change and then identify important gene.