I have never handled PTR-MS data (I have worked with GC-MS data) but I guess your results are a matrix for each sample (time x m/z). If that is the case there are several approaches to build a PCA model, depending on your goals.
If you want to study only one sample, this is, one sample over time, you can just use the time x m/z matrix as X-block for the PCA, where presumably the first PC will be the main drifts of the sample. And the subsequent PCs may explain specific changes in specific time-points.
If you want to study several samples you will get a 3D block so you cannot apply PCA directly. In this case you need to do some kind of unfolding or data transformation. This again depends on what you are aiming for but I could suggest you to average all the mass spectrums over time for a sample and then build a matrix that is sample x m/z, and then apply PCA. However, if you want to compare how the different samples evolve over time, you could take the scores of the first PC (from the previous approach) of each sample and build a matrix that is sample x scores in time, and apply PCA again.
Otherwise you could use multi-way data analysis algorithms such as PARAFAC2 and study everything at the same time, but this is a plan B, I suggest.
I believe this is a really interesting problem so if you want you can give me more insights on the data and I could suggest or try some other things (feel free to email me: [email protected]).
- Badges
- Science topic
- Similar topics
- Biological Science
- Biochemistry
- Enzymology
- Enzymes