Can I use Principal Component Regression/Analysis to identify Raman peaks in varying concentrations? I have data of varying concentrations of a particular substance in water. I recorded its Raman signals between 1 ppt to 1 ppb using identical parameters. 
I would be using Savitzky-Golay filtering to smooth out my data for preprocessing. 
How should I go about using these data to automatically identify the peaks in the spectra?
Can I just perform Principal Component Regression between the spectra and the concentration and look at the loadings? Wouldn't that tell me the wavenumbers that have the highest variations between data points? 
 A: First of all, welcome to cross validated.

Raman signals between 1 ppt to 1 ppb using identical parameters.

Raman or SERS?

How should I go about using these data to automatically identify the peaks in the spectra?

This depends on the precise goal of your data analysis. 
If the main goal is prediction, I'd recommend not to try automatic peak picking but rather go for a method that works well on wider spectral ranges. Principal component regression is a good choice, as is PLS. Ridge regression should work ad well.

Can I just perform Principal Component Regression between the spectra and the concentration and look at the loadings? 

In that case look at both loadings and regression coefficients and at the final coefficients that combine the two.

Wouldn't that tell me the wavenumbers that have the highest variations between data points?

Yes. But you could also use PLS which had latent variables that focus more on variation that helps with the prediction and downweights unrelated variance.
As for the preprocessing, I'd not us Savitzky-Golay smoothing. If you need to smooth, use a technique that allows simultaneously with the smoothing to downsample the spectrum accordingly.
Also, if your knowledge of the analyte and application allows you to further narrow the spectral range, do so. 
