I want to make a PCA analysis with data based on spectra. The PCA is working fine and I also receive nice results. The problem is that my Loadingplot is totally overcrowded because I have around 3500 variables. It would be so much better to display the Loadingplot as a spectrum instead of vectors (in this case the x-axis in the loadingplot shall represent the wavenumber and the y-axis the loading on one PCA).
My data is sorted like in the table below with the difference that I have more than 5 variables and more than 3 samples ;-)
It would be even better if it is also possible to do that with mass spectra. In this case it would be really nice if the data could be sorted after m/z@a certain time. So that 2 information are grouped together. if possible. M/z is in one column and time in another (in the raw data).
I would offer up that you may get something better out of Partial Least Squares analysis. This type of analysis seems to work very well with spectral data. You can learn more about how to use Partial Least Squares via link below.
There is an associated book that gives a very nice example of how to analyze NIR spectra and then predict octane rating.