I’m currently trying to build a models using PLS-DA. I want to examine the regression coefficients for my models and I understand I can extract them from the predict() function. In the output I have coefficients for each component labelled as “dim#, ,”. Should I not have just one coefficient for each feature in my model?

I’m wondering do I take the coefficients from the last dimension as that was the number of components used to create my model? Or should I try to summarize the coefficients across the dimensions?

Because of the iterative nature of the PLS methods, you will have the coefficients for each dimension / component. Assuming you are happy with the choice of the number of components in PLS-DA, then you should have a look at the last component (which aggregates all the information learnt from the previous components). But looking at each component individually (e.g through the predict() outputs $class) will also give you some insights about which groups are discriminated on which components, and what are the variables that enable this (see also plotLoadings(), and looking at the loading values in general with selectVar()).