I have a question concerning principal component analysis and how to pursue.

First of all, this is my case:
I have a lot of data points containing a calculated Q and a measured T.
Q is calculated by the sum of different q's.

I would like to make the following diagonal clusters and trendlines.

I already used PCA to investigate the influence of the different q's on the data.

But I can't figure out how to continue, how can I pursue to make those clusters (and trendlines)?
I'm working in Matlab, but spss or excel (or a combination) is fine as well.

For those who want to know, my final goal is to use those trendlines to predict differences in Q when T would differ with 'the same q-conditions'