PLS-DA vs Multiclass Discriminant Analysis

Hi everyone!

I would like to start by thanking the team behind this wonderful project!

My question is regarding the similarities and differences between Projection to Latent Space (PLS) and Multiclass Discriminant Analysis (MDS), which is an extension of the classical LDA.

They are both supervised learning approaches that take into account the class structure. I think a methodological difference is that PLS is a matrix factorization technique, while MDS is a classifier, but I believe that PLS can serve a similar job, for example, when provided by a new expression profile.
And generally, I’ve noticed that they are always presented as a supervised alternative to PCA.

I’m kick-starting my PhD in integrative analysis and these methods, and likes, are central to my work. As experts in the field, I hope you can help me reach a better understanding of both methods by highlighting the points of differences and similarities.

Thanks in advance!
Mohamed Shoeb

Dear @mohamed.shoeb,

I am not aware of Multiclass Discriminant Analysis. MDS stands for multi dimensional scaling but it is not supervised. The other method is Multi class linear discriminant analysis (LDA).

If you have a relevant publication about MDS, send it here, otherwise you can have a look at this paper where PLS, LDA and CCA are compared: