Depends on the goal. Non-targeted is a discovery/exploration approach, so works for advancing fundamental knowledge of poorly understood conditions/diseases. Targeted requires prior knowledge on the analyte of interest, but is best for quantitative comparison, so it may bee more suitable for well understood clinical presentations.
Elena V Romanova great considerations! Many thanks. I totally agree with your points, but even for very well known diseases, such as Diabetes for instance, we still see a lot of differences among patients, right? So, in that cases wouldn't be interesting to apply a non-targeted approach to better understand the individuality of the patient? The main point to me is the control reference values... a plasma pool? the same individual in different time points?
In my opinion, both unsupervised and supervised metabolomics approaches have their roles to play at individual level. However, i will settle for unsupervised because it is more objective and always an eye opener, especially in pattern recognition and fingerprinting.