I'm evaluating the ability of FT-IR spectroscopy associated to principal component analysis (PCA) to discriminate FTIR data collected from two experimental groups. For this, I'm extracting chemical information from the loadings associated to the PCs that better discriminate both groups.

How many PCs I should consider? And how do I select the better PCs to assess their loadings?

I'm comparing each PC to each other (up to the tenth PC, i.e. PC-1xPC-2; PC-1xPC-3; PC-1xPC-4, ..., PC-9xPC-10) to visualize the data discrimination in the score plots of each pairwise comparison aiming to select the PCs that better discriminate the data and, in the following, to assess their loadings.

Is there some statistical method to select the loadings and PCs that I should consider?